International Journal of Intelligent Systems and Applications in Engineering

Deep Learning Framework for Skeletal Age Classification from Pelvic Radiographs using K-fold Cross Validation and Stacking of CNN Models

S. Jenifer — 2025-04-19

Recent technological developments in deep learning environments have improved bone age evaluation, making it easier and more exact than classic methods in forensic radiology. Deep convolutional neural networks are highly effective at detecting bone age; however, their complexity arises from the number of parameters they need, making them resource-intensive to run on CPUs. To address this, the proposed work utilizes the transfer learning approach to build a two-stage deep learning model based on pelvic radiographs, comprising a vital bone extraction model and an age assessment model. Initially, UNet model combined with Attention Gate extracted the pelvic girdle bones by filtering insignificant regions from pelvis X-rays. For age assessment, a smaller classifier network was first developed and evaluated using K-fold cross-validation. Subsequently, the two deep networks were built by layering the new ones to the existing framework. To enhance performance further, the outputs of both the classifiers were stacked using a dense layer called an aggregator. This meta-learner combined the strength of each model to make decisions on final prediction. The whole framework was validated to analyse its ability to categorize human age in the range of 0–19 years using the collected pelvic radiographs and achieved an average classification accuracy of 97.50%, precision of 98.25%, recall of 96.65%, and F1-score of 97.20%. Thus, the proposed framework can increase the accuracy of multi-classification tasks while leveraging the limited computational resources.

Natural Language processing used in Surgery Implementing with Robot

Nandagopal Redd — 2025-04-19

This article liberalize this machine learning features as it is utilizes within the emerging edge and as feature highlighter to speech recognition approaches on present-day surgical robots. The desire is to advance the event of medical robots among the machine learning and speech recognition liberal that has opened up from the purpose of view of health care services in social protection. The machine learning hypotheses and models are used for pattern recognition structures combined with speech synthesis model with advanced robotic options in medical field. Machine learning is displayed within the comprehension of speech recognition components and its influence in biomedical robots for surgeries. Topical advances of machine learning and intelligent algorithms, further accentuations on their vast hugeness within the improvement of speech recognition in medical surgical applications

Handwritten Text Recognition Using Deep Learning: A CNN-LSTM Approach

Josephine Prem Kumar — 2025-04-19

Handwritten Text Recognition (HTR) has undergone major improvements due to the rise of deep learning. This research introduces a novel approach to hybrid Convolutional Neural Networks (CNNs) in conjunction with Long Short-Term Memory (LSTM) model for accurate recognition of handwritten text. The model is trained using the IAM dataset, consisting of 13,353 handwritten text lines and 115,320 words. The preprocessing pipeline includes grayscale conversion, normalization, and data augmentation to enhance generalization. The CNN is responsible for capturing spatial features from input images, while the LSTM captures sequential dependencies in text, followed by the CTC (Connectionist Temporal Classification) loss function is employed for alignment. Experimental results show an overall Character Error Rate (CER) of 4.57% and a Word Error Rate (WER) of 12.3%. The model outperforms traditional OCR methods and demonstrates robustness in recognizing cursive, printed, and mixed-script handwriting styles. This research highlights the potential of deep learning in real-world used in various applications, including digitizing historical documents, bank cheque processing, and automated postal services.

Optimizing Disaster Management with Blockchain Technology: A Decision Support System for Disaster Risk Reduction and Management

Shyra Mae Eduardo-Igo — 2025-04-19

As technology advances, the enhancement and development of disaster-related remains limited, and disaster management at the regional level is prompt to actively collect and deliver information at a fast pace while deriving comprehensive disaster insights in real time. However, many organizations still rely on manual reporting as it requires formatting, sorting and proofreading that leads to time consuming, data duplication and delays in decision-making and inefficiencies due to lack of appropriate tools to enhance the organization’s productivity. To address these challenges, the researcher developed a Decision Support System with blockchain technology for Disaster Risk Reduction Management for the Office of the Civil Defense Cordillera Administrative Region. This system standardized disaster risk management system and enables for regional agencies to deliver efficiently in near real-time scenario. Additionally, it facilitates streamlined analysis and secure data storage, allowing duty officers to visualize the current situation more effectively. Future researchers can further enhance the system’s functionality by adding recommended features such as AI monitoring and notification, import and export of situational reports from different line agencies, plotting of tropical cyclones related incidents, earthquake and fire incident monitoring.

AI Solution for Lung Cancer Prediction

Leela Prasad Gorrepati — 2025-04-19

Lung cancer is a prevalent chronic condition that is significantly influenced by lifestyle factors, environmental exposures, and genetic predispositions. With smoking being the leading risk factor, habits such as poor diet and lack of physical activity also contribute to the disease's onset and progression [1]. Lung cancer, one of the leading causes of cancer-related deaths worldwide, has a staggering impact on public health, accounting for approximately 25% of all cancer fatalities [2]. In 2022, healthcare expenditures related to lung cancer treatment reached an estimated $18 billion in the United States alone, representing a growing financial burden on both the healthcare system and society. Approximately 230,000 new cases of lung cancer are diagnosed each year in the U.S., with survival rates remaining low, particularly due to late-stage diagnosis [3]. The troubling trend of increasing incidence rates calls for urgent attention, as projections suggest that by 2030, the number of new cases could rise significantly if preventive measures are not implemented. This white paper aims to emphasize the importance of lifestyle modifications, early detection strategies, and public awareness campaigns to mitigate the risks associated with lung cancer [4], ultimately seeking to improve patient outcomes and reduce mortality rates linked to this chronic illness.

ADRC controller for buck converter - inverter system for photovoltaic applications

Esteva Pérez, R — 2025-04-19

This paper presents the design and simulation of the Active Disturbance Rejection Control (ADRC) for a buck-boost converter-inverter system powered by photovoltaic panels. The system features a two-stage configuration, based on a buck-type DC/DC converter and a single-phase inverter, connected in cascade. The objectives of this work are the voltage regulation at the output of the buck converter and the sinusoidal current tracking at the output of the inverter. Active Disturbance Rejection Control is employed from the perspective of differential flatness and Generalized Proportional Integral (GPI) extended state observers. The simulations of the proposed photovoltaic system are carried out in the MATLAB/Simulink environment. Additionally, to evaluate the performance of the photovoltaic system and the robustness of the controller, simulation results are presented under input voltage variations and load disturbances. Furthermore, Total Harmonic Distortion (THD) graphs of the sinusoidal current at the output of the inverter are provided. The simulation results demonstrate appropriate behavior of the Active Disturbance Rejection Control, as evidenced by a short settling time of the buck converter and accurate current reference tracking.

Discovering the Factors Affecting E-CRM Using Machine Learning Techniques

Shymaa Mohamed Mohamed Abdeldayem — 2025-04-19

The majority of business organizations, especially those in developing nations, have adopted E-CRM as a recent strategy, and as a result, managers have strategically invested in online technologies while putting an emphasis on the creation and maintenance of valuable connections with valuable clients. The purpose of this study is to determine the association between E-CRM and service excellence, client satisfaction, loyalty, and trust in Egyptian commercial banks. This study was conducted with the goal of improving E-CRM. For this, 205 valid surveys from bank customers who used E-CRM services were gathered. Data was collected through a survey and utilized for machine learning-based research model assessment (ML). Artificial neural networks (ANN), linear regression models (LRM), random forests (RF), decision trees (DT), K-nearest neighbours (K-NN), and support vector machines (SVM) are examples of machine learning approaches that applied to develop predictive relationship between E-CRM and the other factors. Model performances were evaluated using various statistical indices including the coefficient of determination (R2), Mean square error (MSE), Root mean square error (RMSE), Mean absolute error (MAE) and Mean absolute percentage error (MAPE). The results revealed that E-CRM had a strong effect on service quality, trust, and customer satisfaction while Very Week effect on customer loyalty where The R2 value equal 0.2%.

Reinforcement Learning Hadoop Map Reduce Parameters Optimization

Nandita Yambem — 2025-04-19

Among the various techniques for enhancing Hadoop performance—such as intermediate data compression, in-memory management, and parameter tuning—dynamic configuration parameter tuning proves to be the most impactful. However, existing approaches face several challenges: limited adaptability to specific application requirements, isolated parameter tuning without considering interdependencies, and inaccurate linear assumptions in complex environments. To address these issues, this study introduces a reinforcement learning-based optimization framework using Q-Learning. The proposed method dynamically adjusts key Hadoop configuration parameters by continuously learning from job execution metrics such as completion time and wait times in map/reduce phases. It employs a reward-based feedback mechanism to minimize the gap between expected and actual performance, ensuring more accurate, adaptive, and holistic optimization. Additionally, the framework integrates a neural network to predict optimal parameter values, further enhancing decision-making. This approach significantly improves execution efficiency and resource utilization, offering robust adaptability across diverse workloads and operational environments, while aligning closely with service level agreements.

Grid Quality Enhancement via Direct Power Control in Photo-voltaic-Integrated Energy Systems

Layate Zakaria — 2025-04-19

This paper introduces a Direct Power Control (DPC)-based strategy designed to enhance power quality in grid-connected photovoltaic (PV) systems. The proposed system architecture incorporates a DC/DC boost converter alongside a three-phase inverter, ensuring efficient energy transfer and effective grid interaction. To maximize energy extraction from the PV array, the Perturb and Observe (P&O) algorithm is employed for Maximum Power Point Tracking (MPPT), enabling optimal performance under varying environmental conditions. The DPC methodology directly regulates instantaneous active and reactive power, achieving unity power factor (UPF) operation by maintaining reactive power at zero. The control strategy employs rotating coordinate transformations and precise grid voltage phase tracking, resulting in robust synchronization and improved dynamic behavior. To further simplify the design and improve the grid performances, an LCL filtering stage has been introduced in this work. Hence, the proposed approach demonstrates significant robustness against load disturbances and variability in operating conditions. Through detailed simulations, the system's performance is validated, showcasing its ability to enhance grid quality, maintain stability, and ensure seamless operation during both transient and steady-state scenarios. The results emphasize the potential of this DPC-based control strategy as an efficient, reliable, and scalable solution for improving grid power quality in PV-integrated energy systems.

Transparent Floating Solar Cells: Advancing Renewable Energy with Dual-Use Technology

Wahiba Slimani — 2025-04-19

The emergence of transparent solar cells in floating stations is an original concept. It can produce energy without requiring significant, substantial land, and it helps prevent evaporation, which results in the loss of between 50% and 70% of water. Their continuous cooling benefits, weight, and high absorption range favor their use. Transparent floating solar panels are also favored for their lowest cost. Many studies are focused on their potential dual-use applications, cost-effective manufacturing methods, and promising raw materials for this technology. This paper surveys suitable transparent solar cells that can be used in floating stations instead of silicon-based solar cells according to their features, adaptability, and efficiency, which can sometimes surpass conventional solar cells.

Explainability Measurement of Machine Learning Model in Phishing Detection

Abdullah Fajar — 2025-04-19

Explainability in phishing detection models can enhance phishing assault mitigation by fostering confidence and elucidating the detection process. The essential requirements for facilitating human comprehension and assessment of the reasons a specific URL is deemed insecure for visitation. The aims of this study are to investigate some machine learning models in phishing detection which have abilities to fulfil the critical needs of explanation using explainability metric. This study applies a methodology starting with dataset collection of phishing and legitimate URL as the sources of various features. Then the models selected, which are often known have good quality in classification between phishing or legitimate label. The modeling results are processed using an explainer method to generate a comprehensive understanding of feature behaviors that influence model predictions. Instead of present accuracy metric results only, this study discusses how explainability metric shows how the features contribute to the model. The conclusion shows that some features have abilities to influence the model decision in general or specifically, then how the features contribute to the model in terms of stability and distribution behaviors. The study shows that some features that may be identified as key features of model behavior then can be applied practically to phishing detection systems such as firewall or SIEM (Security Information and Event Management).

Retracted

Retracted — 2025-04-19

Retracted

Hybrid Reinforced Pelican Optimization Algorithm (HR-POA) for Energy-Efficient Cluster Head Selection in Heterogeneous Wireless Sensor Networks

Maroua Hammadi — 2025-04-19

The importance of Wireless Sensor Networks (WSNs) spans multiple areas, notably environmental monitoring, healthcare services, and intelligent urban systems. These networks consist of dispersed sensor nodes that wirelessly exchange data while monitoring environmental or physical conditions. One of the key challenges in Wireless Sensor Networks (WSNs) is the efficient selection of Cluster Heads (CHs) to prolong network lifetime and ensure balanced energy consumption. This study introduces a novel Hybrid Reinforced Pelican Optimization Algorithm (HR-POA), which integrates the Enhanced Pelican Optimization Algorithm (EPOA) with Particle Swarm Optimization (PSO) and Reinforcement Learning (RL) to achieve efficient Cluster Head (CH) selection in heterogeneous wireless sensor networks (HWSNs). HR-POA is a promising solution for energy-efficient clustering since it considerably improves WSN performance by utilizing intelligent routing and a hybrid optimization approach. The proposed algorithm considers node energy, distance, and adaptive Q-learning-based routing to improve energy efficiency and network performance. The effectiveness of HR-POA in comparison to current CH selection algorithms has been assessed through extensive simulated studies. HR-POA demonstrates notable improvements in energy efficiency, network longevity, and packet delivery ratio compared to current CH selection methods, as evidenced by simulation results. By advancing energy-aware clustering approaches in WSNs, the suggested approach opens the door to more intelligent and sustainable wireless sensor networks.

A Comparative Analysis of AI-Based Chatbots for Disease Diagnosis Based on Symptoms

Leela Prasad Gorrepati — 2025-04-19

Investment in artificial intelligence (AI) chatbots within the United States has experienced substantial growth in recent years, driven by the increasing demand for intelligent virtual assistants across various sectors. By 2025, the United States is anticipated to be a significant contributor to the global AI chatbot market, which is projected to expand from $8.6 billion in 2024 to $11.14 billion in 2025 [1]. This growth represents a compound annual growth rate (CAGR) of 29.5% [1]. The expansion is primarily attributed to the rising demand for automated customer support, personalized digital experiences, and the widespread implementation of AI technologies. Major technology companies and venture capital firms are making noteworthy investments in the development of AI chatbots. For instance, Gloo, a technology firm focused on faith-based solutions, raised $110 million to advance AI tools [2], including chatbots. Concurrently, Yutori, a startup founded by former executives from Meta AI, secured $15 million to develop sophisticated AI personal assistants [3]. On a broader scale, prominent U.S. technology firms, including Microsoft, Amazon, and Google, are investing hundreds of billions of dollars into AI infrastructure and applications, with a focus on chatbot technologies. Microsoft, for instance, has outlined a plan to allocate $80 billion to AI initiatives in 2025 [4], while Amazon has invested $8 billion in Anthropic, a leading AI startup specializing in generative AI and chatbot solutions [5]. These investments underscore the strategic significance of AI chatbots in enhancing customer engagement, optimizing operations, and fostering innovation across various industries, such as healthcare, finance, retail, and education. As AI capabilities continue to advance, the United States is poised to maintain its leadership in chatbot innovation and commercialization. Artificial Intelligence (AI)-driven chatbots have emerged as transformative tools within the healthcare sector, playing an increasingly vital role in facilitating symptom-based disease diagnosis. These intelligent systems harness a suite of advanced AI technologies—including machine learning, natural language processing (NLP), and knowledge representation frameworks—to interpret patient-reported symptoms, assess potential health conditions, and provide users with preliminary diagnostic insights [6], [7].This white paper offers a comparative evaluation of the capabilities, strengths, limitations, and practical applications of leading AI-powered chatbots specifically designed for disease identification. It explores the algorithms and datasets that drive their diagnostic reasoning, the accuracy and reliability of their outputs, and their ability to adapt across varying clinical scenarios and patient populations [8]. The paper highlights the critical role of AI chatbots in expanding access to healthcare, particularly for underserved or remote communities where traditional medical resources may be limited [9]. By offering on-demand symptom assessment, triage guidance, and referral recommendations, these systems empower individuals to seek timely medical attention, potentially reducing the burden on emergency departments and primary care providers. In addition to assessing current implementations, this paper addresses key challenges—including data privacy, diagnostic accuracy, regulatory compliance, and patient trust—that impact the broader adoption and integration of AI chatbots in clinical practice [10]. It explores how these technologies could evolve to support predictive analytics, personalized medicine, and integrated care pathways, ultimately contributing to more efficient, responsive, and patient-centered healthcare systems [11].

Web-Based Solution for Efficient Waste Management

Pooja Sharma — 2025-04-19

Electronic waste (e-waste) is a rapidly growing worldwide environmental problem because of technological advances, reducing product lifespans, and big consumer demand [1],[2].E-waste contains vital substances such as gold, silver, and copper, as well as unsafe substances including lead, mercury, and cadmium, that could cause brilliant damage to human fitness and the surroundings if now not managed correctly[1].

Effective e-waste solutions consist of many techniques, which include recycling, reuse, sustainable product improvement, and public focus [3]. Methods to support the improvement of a round economic system and renewal, and to put into effect extended manufacturer responsibility (EPR) guidelines to hold groups accountable for the control of give up products. Further, new technology together with advanced analytical techniques and green extraction technologies can increase the recycling fee[2],[4].

Public schooling programs and worldwide cooperation are important to remedy the e-waste trouble and decrease its worldwide harm.

The need for cooperation between authorities, enterprise, and consumers to create a smooth, in experienced future.

Mapbi3 Perovskite Solar Structure Experimental Optimization: Control of Deposition of Spinning Speed (Comparative Study of Optical and Structural Proprieties)

Aicha Aziza Ayad — 2025-04-19

The present paper deals with the stability of halide hybrid perovskite MAPbI3 (CH3NH3PbI3) thin films by an experimental optimization by increasing the Dimethyl sulfoxide (DMSO) solvent ratio (compared to literature) in perovskite solution and the study of deposition spinning speed effect on the structural and optical properties. The samples is deposited by a spin coater in one step with 400, 1500 and 3000 rpm spinning speed. Structural and optical properties of the elaborated samples were studied using X-ray diffraction and UV-Visible spectroscopy. The main results show that compared with other works the DMSO solvent ration raise has a positive effect in the perovskite proprieties, in other hand, the best proprieties of perovskite need an optimal speed deposition; the very high or low speed has a negative effect.

VoiText Care: ANFIS driven Persuasive Mobile Application for Medication Adherence Intervention

Uzoma R Alo — 2025-04-19

The prevalence of medication non-adherence to long-term therapies among outpatients with chronic disease has continued to be an issue of serious concern to healthcare institutions and general public. Multifaceted intervention approach to motivate and promote positive health behavior of patients towards medication adherence is highly needed. In this paper, the design of voice and text (VoiText Care App, a persuasive mobile application that leverages patient’s assessment score of medication non-adherence level (data) with four (4) linguistic terms (variables) generated by ANFIS algorithm for delivering of personalized and persuasive adherence intervention message is proposed. The four linguistic variables (terms) of medication non-adherence level are: very low non-adherence (VLNA), low non-adherence (LNA), high non-adherence (HNA) and very high non-adherence (VHNA). With assessment score, level of non-adherence of patient is mapped with linguistic terms and it is used to determine the persuasion strategy or principle to be adopted for the composition of the persuasive messages. With combined potential of agent voice call, short message service of mobile phone technology, persuasive strategies and web portal, the developed persuasive mobile application could efficiently improve adherence to medication.

The Challenges of Machine Translation from Telugu Literary Text to English: Textual and Contextual Analysis

Kolakaluri Madhu Jyoti — 2025-04-19

Translation from one language to another is both an art and science. Machine Translation (MT) has been in existence since 1940s and has flourished in recent times due to the proliferation of the web. MT plays predominant role by enabling faster access in getting required translation. However, it is difficult to accept the reality that translation is effective when it comes to literature. Literature is not just a text; it is a collection of expressions, emotions and feelings. Although artificial intelligence is a lead for MT and referred as an intrinsic human quality, it is seldom capable of defining or analysing. Consequently, machine translation particularly for literature facess several challenges. The current paper aims to explore the challenges of MT as an aid used to translate Telugu modern short fiction (Source Language) into English language (Target Language) by online Google Translator (GT). Similarly, the study focusses on various types of language efficacies and textual related hitches such as: semantics, syntax, cohesion and coherence followed by semantics extractions and discourse resolutions. This analysis delves critically into the specific hurdles faced by MT in the context of parsing difference between standard text and machine generated translation.

The Deep Learning Approach for Crop Selection and Yield Prediction using Bi-LSTM in Agriculture

M. Supriya — 2025-04-19

Accurate crop yield prediction is essential for optimizing agricultural practices, ensuring food security, and maximizing resource efficiency. Traditional methods often fail to capture the complex, sequential dependencies in agricultural data, limiting their predictive accuracy. This work focuses on improving crop yield prediction by overcoming the drawbacks of conventional methods and integrating sequential data. The presented Bi-LSTM model provides better results than other machine learning and deep learning models since it uses all dependencies of temporal data of agriculture data. The study used Agricultural Crop Yield dataset then training and testing Bi-LSTM model. The performance is compared with other methods such as Linear Regression, Random Forest and basic LSTM to determine Mean Absolute Error, Root Mean Squared Error, R² score and Mean Absolute Percentage Error. The Bi- LSTM model yields the best result with MAE=0.32, RMSE =0.47 and R² Score =0.91. It efficiently incorporates features like rainfall, usage of fertilizers, which proves its applicability in the data of crop yields data. The analysis proves Bi-LSTM to be effective in predicting crop yield and offers a sound approach for decision support in agriculture.

Visual Intelligence: A Triple-Attention Network for Robust Fall Detection in Complex Environments

Nawaf A. Alqwaifly — 2025-04-19

Real-time fall detection in complex environments remains a challenging task due to varying human postures, occlusions, and cluttered scenes. This paper presents Symmetry-Aware Visual Intelligence, a novel triple-attention network built upon an enhanced YOLOv5 backbone to ensure robust detection without sacrificing computational efficiency. Our approach integrates three complementary attention mechanisms: Local Attention in early convolutional layers to emphasize posture-relevant spatial symmetry, Squeeze-and-Excitation (SE) blocks within the backbone to recalibrate channel-wise feature importance, and Efficient Channel Attention (ECA) in the neck for improved multi-scale feature fusion. Together, these modules enhance both spatial precision and contextual awareness. The proposed architecture achieves state-of-the-art results, with mAP scores of 0.914 on the DiverseFall dataset and 0.994 on CAUCAFall, outperforming baseline YOLOv5s by 7.7% and 8.2%, respectively. Notably, it also surpasses YOLOv5x in precision (0.903 vs. 0.769) while maintaining a lightweight design with 80% fewer parameters. Extensive ablation studies validate the contribution of each attention module, and training optimization using SGD at a learning rate of 0.001 ensures convergence. Our model offers a high-performance, efficient solution for fall detection in real-world scenarios with structural complexity.

Intersection of AI and Cybersecurity: A Data-Driven Approach to Proactive Risk Management in ETL Processes

Shiva Kumar Vuppala — 2025-04-19

The growing complexity of Extract, Transform, Load (ETL) processes and their crucial role in modern data pipelines make them susceptible to various cybersecurity risks, including unauthorized access, data tampering, and service disruption. These threats can have far-reaching consequences, affecting business operations, regulatory compliance, and strategic decision-making. Traditional security approaches, relying on static rule-based systems, struggle to address the dynamic nature and scale of ETL workflows, necessitating the integration of more adaptive and intelligent methods. A data-driven approach utilizing Artificial Intelligence (AI) offers a promising solution by leveraging machine learning and deep learning techniques to continuously analyze system logs, performance metrics, and historical incidents for abnormal activity. This paper proposes a hybrid approach combining autoencoders for feature extraction and Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) models for anomaly detection, aiming to proactively identify security risks within ETL systems. Autoencoders are employed to reduce data dimensionality while capturing critical features, while the CNN-GRU model enhances the detection of both local and temporal anomalies. The proposed method is evaluated through performance metrics, showing a high detection rate and minimal false positives compared to traditional rule-based methods. The results demonstrate the potential of AI-driven security frameworks to provide real-time, intelligent monitoring and adaptive risk management, thus improving ETL pipeline resilience and security. This research highlights the importance of incorporating AI into cybersecurity strategies for dynamic, data-intensive environments, ensuring that security measures evolve alongside emerging threats.

Compression of Visual Images based on Histograms using Quadtree Algorithms

Nawaf A. Alqwaifly — 2025-04-19

Over the last few years, there has been an exponential growth in the demand for images and video sequences via wireless networks. The result has been that image and video compression has become an increasingly crucial issue in decreasing the cost of storing and transmitting data. The goal of visual image compression is to reduce the amount of information required to represent an image. To compress an image efficiently, a technique is used to reduce the space required and to increase the efficiency of transferring the image over the network in order to improve access to the images. In this paper, we present a histogram-based visual image compression technique based on Quadtrees. Using this technique, the image is divided into blocks in order to reduce the space necessary for the whole image. This ensures the efficient transmission of each block. A histogram of the image block is used to analyse the compression of an image. The results of the experiments indicate that the algorithms provide a compression ratio that varies between 0.13 and 0.61. Moreover, the results prove that the method is able to improve the compression performance and can achieve a similarity between the compression ratio and image quality.

Application of 3D Laser Scanner in Digitization and Virtual Reality of Numido-Punic Funerary Monument: The Soumaa's Mausoleum (Constantine, Algeria)

Benzagouta Yasser Nassim — 2025-04-19

Algeria's cultural heritage, from ancient structures to stunning landscapes, serves as a powerful symbol of both its national identity and humanity's shared knowledge. The Soumaa of El Khroub, a Numido-Punic mausoleum in north-east Algeria, dating from the 2nd century BC, it was classified as a national monument in 1900 and recognized by UNESCO in 2002. However, despite its inestimable value, this heritage faces a number of risks, whether natural, environmental or anthropogenic. Although institutions recognize its importance, lack of proper maintenance and harmful actions are contributing to irreversible damage. The aim of this research is to develop a digital documentation, visualization of a funerary monument and data management. monument through the creation of a detailed 3D model. This involves employing laser scanning TLS 3D and point cloud processing to obtain precise measurements of the monument's components, aiming at facilitating restoration and preservation efforts. It will be also accessible to researchers, curators, architects and other professionals. To cover the entire site, 17 scanning stations have been planned, including 13 for the monument itself and 4 for the site as a whole. The end result will be a three-dimensional representation of the monument with all its dimensional details, incorporating a variety of data such as structural, architectural, historical and technical aspects. In addition, to enhance the cultural importance of the monument, the data from the 360-degree scan was used to create a virtual reality experience, enriched with descriptive texts and photographs. This initiative demonstrates a positive commitment to the use of virtual reality as an educational tool, offering visitors an interactive immersion in the history and significance of the monument. This innovative also makes it accessible to a wider audience, fostering a greater understanding and appreciation of its historical and cultural significance.

Optimal Super-Twisting SMC Design for CSTR via Improved Grey Wolf Optimization and Digital Implementation

Hammad Iqbal Sherazi — 2025-04-19

The Continuous Stirred Tank Reactor (CSTR) is a widely studied system in the field of process control due to its nonlinear and time-varying behavior. Characterized by second-order nonlinear dynamics, it poses significant challenges in maintaining system stability and performance under varying operating conditions. Owing to these complexities, the CSTR is frequently employed as a benchmark model for evaluating the efficacy of modern control techniques. In this research, a condition-based Super-Twisting Sliding Mode Controller (STSMC) is developed to enhance the robust- ness and accuracy of the control system. The controller is specifically designed to handle the nonlinearities and external disturbances inherent to the CSTR process. A comprehensive stability analysis of the proposed control scheme is carried out using Lyapunov stability theory, ensuring that the system trajectories remain bounded and converge to the desired equilibrium. To further improve the control performance, the gains of the STSMC are optimally tuned using an Improved Grey Wolf Optimization (IGWO) algorithm. This metaheuristic optimization technique is employed to achieve faster convergence, better tracking performance, and reduced steady-state error. The complete control architecture is then implemented and validated on a Delfino C2000 digital controller to evaluate its real-time performance. Experimental results confirm the practical applicability and effectiveness of the proposed method in achieving stable and robust control of the CSTR system.

Enhancing Tax Administration in Niger : A Data Mining Approach to Outlier Detection

Moussa Khane — 2025-04-19

Developing countries face significant challenges in accurately forecasting tax revenues due to disparate databases and the presence of outliers in collected taxes. These anomalies can lead to inconsistencies in revenue predictions, impacting economic planning and policy decisions. This study applies the CRoss Industry Standard Process for Data Mining (CRISP-DM) framework to support Niger’s tax administration in detecting and addressing outliers. Boxplot analysis and extreme value detection algorithms were utilized to visualize outliers, while the Interquartile Range (IQR) Machine Learning (ML) algorithm was employed to remove them. The dataset covers the period from January 2019 to December 2022. The current analysis identified significant outliers in June 2020 and December 2021 for Value Added Tax (VAT) and in August 2021 for Processing Tax and Salary (ITS). The study found that with outliers, VAT, ITS, and Profit Tax (ISB) accounted for 61.2% of total tax revenues, whereas without outliers, their combined contribution increased to 64.8%, highlighting the importance of accurate anomaly detection for better fiscal planning.

Exploring the Synergy of Generative AI and Large Language Models Advancing Machine Learning Applications in Data-Driven Research

Krishnam Raju Narsepalle — 2025-04-19

Generative AI and the large language models (LLMs) are powerful new components in ML, and platforms capable of supporting these technologies deliver remarkably sophisticated data-driven applications. This paper explores the joint application of such technologies along with its potential of enhances other machine learning implementations. A detailed exploration of how generative AI models like GANs and diffusion models, converge with LLMs to solve both natural language processing and multimodal data synthesis problems are revealed through this paper. Our empirical evidence illustrates how the co-deployment of generative AI models and LLMs is shown to improve performance by augmenting data scenarios as well as applying an integrated approach to context retrieval and prediction model accuracy. Our technical approach provides a new framework that integrates generative modeling with LLMs and aims to accelerate research pipelines mainly involving biomedical data analysis and knowledge discovery tasks. Our study shows that this combination will be fundamental reconfiguration of new paradigm of machine learning to provide more robust and advanced scale systems with intelligence. In short, we need generative AI with LLMs to create our strong foundation to build data-driven innovations on top of as we enter different sectors.

Enhancing Sustainable Healthcare: Machine Learning-Based Tuberculosis Detection Using C4.5 Decision Tree

Helanmary M Sunny — 2025-04-19

Tuberculosis (TB) remains a global health crisis, particularly in resource-limited regions where diagnostic infrastructure is scarce. While deep learning models dominate recent research, classical machine learning (ML) methods offer interpretability and computational efficiency—critical for low-resource settings. This study presents the first systematic comparison of 13 ML algorithms, including C4.5 decision trees, logistic regression, and ensemble methods, for TB detection using the Shenzhen chest X-ray dataset .The C4.5 decision tree achieved near perfect accuracy (99.78%) and the lowest training time (0.147s), outperforming deep learning alternatives in interpretability and cost-effectiveness. By providing a deployable, low-cost diagnostic tool, this work directly supports the United Nations’ Sustainable Development Goals (SDGs): SDG-3 (reducing TB mortality), SDG-9 (fostering diagnostic innovation), and SDG-10 (bridging healthcare disparities). Our results demonstrate that classical ML can rival complex models in medical diagnostics while remaining accessible to underserved populations

Real-Time AI-Driven Bug De-duplication and Solution Tagging Using Graph Neural Networks

Alex Thomas Thomas — 2025-04-19

The process of bug tracking and resolution is a critical aspect of software development, yet it is often afflicted by redundancy and inefficiency, especially due to duplicate bug reports and inconsistent solution tagging. This review sees recent advances in AI-driven techniques, especially those utilizing Graph Neural Networks (GNNs), for real-time bug de-duplication and automated solution tagging. I investigate how the relational structures are essential in bug reports and historical fixes can be modeled using GNNs to improve bug triage processes. The review incorporates key methodologies, compares performance across multiple benchmarks, and highlights the benefits and limitations of GNN-based approaches like traditional machine learning and NLP methods. Furthermore, I analyze the combination of such models in real-world development pipelines and discuss their potential to reduce manual effort, advance in debugging workflows, and improve overall software quality. Finally, the paper recognizes open challenges and future research directions, including scalability, real-time inference, and domain adaptation, to guide future innovation in automated bug management.

Design and Evaluation of a 4-Way Oracle GoldenGate Replication Architecture for Hybrid Cloud Environments

Karuppusamy Gopalan — 2025-04-19

In the era of hybrid cloud adoption, enterprises face significant challenges in maintaining data consistency, high availability, and operational continuity across geographically dispersed and heterogeneous environments. This study presents a novel implementation of 4-way active-active Oracle GoldenGate replication across on-premises and multi-cloud platforms, enabling real-time data synchronization, conflict resolution, and automated failover mechanisms [1][2]. The architecture interconnects four nodes—two on-premises and two cloud-based (Azure)—to deliver seamless data replication with less lag, ensuring transactional integrity and zero-downtime operations for critical enterprise applications. Key innovations include the use of Oracle GoldenGate Microservices Architecture, integration with cloud-native services [5], and deployment of Conflict Detection and Resolution (CDR) techniques to handle bi-directional updates in a multi-master configuration [2][3]. Experimental evaluation demonstrates that the proposed system achieves high throughput, minimal replication lag, and robust failover capabilities under diverse workloads and simulated failure conditions [6][7]. The results validate the effectiveness of the 4-way replication strategy in supporting disaster recovery, data sovereignty compliance, and cross-region data access in a hybrid cloud setting. This work contributes to the field of cloud database management by offering a scalable and resilient model for enterprise-grade data replication, applicable to sectors demanding high availability such as healthcare, finance, and e-commerce.

AI Enhanced BI for Dynamic Portfolio Optimization under Market Volatility

Rajesh Aakula — 2025-04-19

Volatile markets make it hard to invest because the values keep fluctuating and there is uncertainty in the market. This is because strategic investment principles may not be suitable to make high-speed movements that may bring low returns or increase risk. BI integrated with Artificial Intelligence (AI) has the benefit of real-time analysis and dynamically designed decision tools that provide constant portfolio fine-tuning. This is an ability offered by AI that involves analyzing high levels of data, detecting market trends, and controlling assets to keep the portfolio on course towards its goal amidst fluctuation. This integration makes dynamic rebalancing possible such that the portfolio is realigned back to real market terms to reduce risk and maximize return. With these technologies, portfolio managers have an easy time coping with risks in the market and offering better investment propositions that are dynamic and technology reliant. Combining artificial Intelligence and Business Intelligence is an efficient way to maximize portfolio management in the modern and uncertain financial environment.

Retracted

Retracted — 2025-04-19

Retracted

AI and ML Algorithms in Cyber Security

Chandrababu C Nallapareddy — 2025-04-19

The rapid evolution of cyber threats, coupled with the increasing complexity of digital ecosystems, has necessitated more intelligent and adaptive security solutions. Artificial Intelligence (AI) and Machine Learning (ML) have emerged as transformative technologies in the cybersecurity landscape, enabling organizations to proactively detect, prevent, and respond to malicious activities with greater speed and precision. This paper explores the integration of AI and ML algorithms in various cybersecurity applications, including threat detection, incident response, vulnerability management, and user behavior analytics. It also examines the alignment of these technologies with established cybersecurity frameworks and standards such as NIST CSF, ISO/IEC 27001, and the NIST AI Risk Management Framework to ensure ethical, secure, and effective implementation. By evaluating real-world use cases and current challenges, the paper underscores the critical role of AI/ML in building resilient, future-ready cyber defense strategies.

AI-Based Prediction of Scope Creep in Agile Projects

Alex Thomas Thomas — 2025-04-19

Scope creep is continuing to be one of the main challenges for Agile software development, usually preventing the projects from being timely completed, as well as causing them to go over budget and reduce product quality. New advances in artificial intelligence (AI) and machine learning have shown promise in forecasting and managing scope creep by monitoring project data, team actions, and risk factors. This document presents a thorough summary of AI usage to scope creep prediction in Agile projects. We comprehensively examine existing machine learning models for effort estimation, risk management, and scope management and uncover methods such as neural networks, deep learning, and ensemble learning. The examination consolidates findings of prominent studies in Agile project risk prediction, automated scope creep detection, and AI-enhanced scheduling and presents their efficacy and limitations. Besides, we identify challenges in the implementation of AI in Agile methods and propose future research areas to increase prediction accuracy and deployment in practice. This survey aims to provide researchers and practitioners with a shared understanding of AI usage in preventing scope creep, thus enhancing Agile project success.

AI-Driven Design Verification of Semiconductor ICs for Graphics Processing Unit Using LLMs

Nilesh Patel — 2025-04-19

The exponential growth of next-generation GPU technologies, for gaming and now AI processing, demands highly reliable and efficient semiconductor chip designs. As chip complexity surges, traditional verification methodologies are increasingly challenged by limitations in scalability, time, and coverage. In this context, Artificial Intelligence (AI), particularly Large Language Models (LLMs), offers transformative potential in automating and accelerating the chip design verification process. This paper presents an AI-driven framework leveraging LLMs for the verification of semiconductor chips tailored for GPU systems. We explore how LLMs can interpret design specifications, generate test cases, identify anomalies, and assist in natural language debugging, thereby significantly enhancing verification throughput and accuracy. The proposed approach integrates LLMs with formal verification tools and simulation environments, enabling contextual understanding of hardware description languages (HDLs) and streamlining functional and system-level validation. Additionally, we examine case studies demonstrating improvements in error detection, coverage analysis, and design cycle reduction GPU components. Our findings show that LLM-assisted verification achieves notable gains in identifying logic bugs, reducing verification effort, and ensuring standards compliance in complex chip designs. We also discuss the challenges of domain adaptation, model fine-tuning for HDL context, and handling proprietary IP sensitivity. Finally, this research lays the groundwork for broader adoption of AI-augmented verification pipelines in semiconductor development for advanced communication technologies. The integration of LLMs into chip design workflows not only enhances productivity but also redefines the paradigm of intelligent design verification, aligning with the rapid pace of innovation in the GPU landscape.

Retracted

Retracted — 2025-04-19

Retracted

Numerical Investigation of Flat Plate Solar Collector by Using Various Types of Mono and Hybrid Nanofluids

Karrar A. Al-shibli — 2025-04-19

The performance of traditional fluids in heat transmission might be greatly enhanced by using nanofluid, a cutting-edge fluid. Improving the design elements and the convection heat transfer coefficient between the fluid and absorber tubes are the most important ways to raise the solar collector’s overall efficiency. In solar collectors, water nanofluids such as MWCNT, Al₂O₃, TiO₂, SiO₂, and CuO are most commonly used nanofluids. This study used controlled conditions to numerically examine the thermal efficiency of flat plate solar collectors using Al₂O₃, MWCNT, and hybrid Al₂O₃+MWCNT (80:20%) as a working fluid. The efficiency is examined in relation to a number of characteristics, such as the volumetric flow rate and the volume percentage of nanoparticles, and the intensity of solar radiation. Six concentrations ratio of different nanoparticles were used (0, 0.01, 0.02, 0.03, 0.04, and 0.05) during this numerical investigation, with each of these concentrations six different mass flow rate were used (0.004167, 0.08334, 0.0125, 0.01667, 0.03334, and 0.05) kg/s. The results showed that the highest efficiency was obtained from using MWCNT/water as a nanofluid (76.8%, 78.3%, and 80.4%) at mass flowrate 0.05 kg/s for concentration ratios (0.03,0.04, and 0.05) respectively. While the base fluid (water) at the same mass flow achieved lower efficiency 53.4%. Al₂O₃/water achieved median efficiency (59.1%, 60.1%, and 61.1%) at mass flowrate 0.05 kg/s for concentration ratios (0.03,0.04, and 0.05), respectively. The efficiency of hybrid Al₂O₃+MWCNT (80:20%) at mass flowrate 0.05kg/s were 75.2%, 76.5%, and 78.3% for concentration ratios (0.03,0.04, and 0.05) respectively.

MiniMind-Dense: a Small Language Model that Supports HR Services by Adopting MiniMind with Supervised Fine-Tuning and LoRA

Darren Chai Xin Lun — 2025-04-19

Small and medium-sized enterprises (SMEs) form the backbone of Malaysia’s economy, yet they often lack resources to adopt advanced AI tools. Generative AI and large language models (LLMs) promise to transform human resources (HR) by automating tasks like policy guidance, recruitment support, and employee coaching [1]. However, generic LLMs trained on broad data do not capture local legal norms or HR-specific knowledge and may produce irrelevant or non-compliant advice. To address this, we adopted and adapted MiniMind [2] , a Malaysian HR-focused language assistant. Starting from a compact GPT-style base model, we implemented a three-stage refinement pipeline: (1) Supervised fine-tuning (SFT) on an expanded, HR-domain dialogue dataset; (2) Reinforcement learning from human feedback (RLHF) via Proximal Policy Optimization to align outputs with HR professionals’ preferences; and (3) LoRA parameter-efficient tuning to inject final domain expertise. Through the refinement pipeline, We have designed a new architecture, namely, “MiniMind-Dense”, by incorporating other Transformer improvements such as grouped-query attention, rotary embeddings, SwiGLU to achieve the goals of the research. Extensive evaluation on Malaysian HR queries shows dramatic improvements: e.g., BLEU score [3] rose from ~5% in the pretrained model to ~79% after LoRA tuning (and ROUGE‑L [4] from ~36% to ~92%). Qualitative analysis confirms highly fluent, relevant, and human-like responses, unlike the generic outputs of the base model. These results demonstrate the feasibility of a localized, aligned SLM as an AI HR assistant. Future work will integrate retrieval (RAG) [5] for factual grounding and expand multilingual capabilities.

Development Of A 33 Ghz Low Noise Amplifier Utilizing The Multigate Technique For Cascode Devices

P. Venkateswarlu — 2025-07-07

The presence of increased parasitic components in silicon-based nanometer (nm) scale active devices results in various performance trade-offs when optimizing key parameters, such as maximum frequency of oscillation (fmax), gate resistance, and capacitance, among others. A common-source cascode device is frequently employed in amplifier designs operating at RF/millimeter-wave (mmWave) frequencies. Besides intrinsic parasitic components, extrinsic components arising from wiring and layout effects are also vital for the performance and precise modeling of these devices. This study presents a comparison of two distinct layout techniques for cascode devices aimed at optimizing extrinsic parasitic elements, including gate resistance. A multi-gate or multi-port layout technique has been proposed to optimize gate resistance (rg). Measurement results indicate a 10% reduction in rg for the multi-gate layout technique when compared to a conventional gate-above-device layout for cascode devices. Nevertheless, the conventional layout demonstrates smaller gate-to-source and gate-to-drain capacitances, resulting in enhanced performance regarding speed, specifically fmax. An LNA has been designed to operate at a frequency of 33 GHz utilizing the proposed multi-gate cascode device. The LNA achieves a measured peak gain of 10.2 dB and a noise figure of 4.2 dB at 33 GHz. All structures have been designed and fabricated using 45 nm CMOS silicon on insulator (SOI) technolog.

Enhancing Healthcare Analytics with Federated Learning and Cloud Technologies for Improved Patient Outcomes

Sreepal Reddy Bolla — 2025-04-19

The rapid health system digitization leads to significant accumulation of patient data that sophisticated analytical tools help doctors improve diagnosis accuracy and treatment decisions without sacrificing treatment outcome quality. Traditional centralized systems prevent military-grade machine learning models that handle healthcare analytics from implementation because of privacy regulations and security concerns coupled with regulatory requirements. FL operates as an appealing decentralized structure enabling institutions to develop their models jointly without needing real patient information transfer throughout collaborative training procedures. FL utilizes cloud systems and AI and data mining to develop predictive healthcare analytics which supports patient privacy standards while meeting HIPAA and GDPR requirements in healthcare. This paper studies the healthcare analytics system improvement processes achieved through combining Federated Learning with Cloud Computing and AI-driven Data Mining. This examination describes the cooperation between Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks and Transformer-based models to enhance medical picture evaluation and disease manifestation and unique treatment solution forecasting within decentralized networks. SMPC techniques together with differential privacy protocols serve as the central aspect of the study to resolve security and privacy constraints in FL system deployments. The research team will optimize healthcare federation networks through blockchain addition while developing FL architectures and improving network communication systems.

Token Bucket: Architecting Next-Gen Secure, Resilient, and High-Performance Distributed Systems and App Modernization

Nitin Gupta — 2025-04-19

With the increase demand of digital footprint, creating a robust and resilient distributed system adhering to cloud needs has become essential. Distributed system has challenges of scaling and cascade failure in case of exponential or burst traffic. Token Bucket algorithm is an architecture paradigm in distributed system that not only helps to overcome these challenges but also to sustain increasing workloads. In digital eco-system Token Bucket helps systems to be loosely coupled with asynchronous processing that serves to design a system with maximum availability and minimum latency. Along with its self-healing and fault tolerance capability Token Bucket also helps in safeguarding the system from various cyber threats. Token bucket provides a blueprint to architect complex software system that delivers speed, reliability and fortified protection. This algorithm when properly designed and considered for use case can be a state-of-the-art for digital ecosystem.

The Role of AI in Strengthening Cybersecurity for Data Pipelines and ETL Systems

Manohar Reddy Sokkula — 2025-04-19

In the era of big data and cloud-native architectures, Extract, Transform, Load (ETL) systems and data pipelines form the core of enterprise-level data processing and decision-making. However, their growing complexity, distributed nature, and continuous data movement have also made them prime targets for sophisticated cyberattacks. Traditional security methods such as firewalls, rule-based monitoring, and static encryption often fall short in identifying evolving threats within these dynamic environments. This research explores the integration of Artificial Intelligence (AI), particularly deep learning models, to enhance the cybersecurity posture of ETL systems. The study presents a hybrid Autoencoder-LSTM-based anomaly detection model designed to monitor and secure ETL workflows in real-time. The model is trained using a combination of real-world network intrusion datasets such as CICIDS2018 and UNSW-NB15, along with synthetic ETL telemetry logs generated through tools like Apache NiFi and Talend. Before model training, data preprocessing using Min-Max normalization ensures consistency and efficient learning across diverse feature sets. Additionally, visual tools such as reconstruction error graphs, threshold-based detection plots, correlation heatmaps, and log activity timelines were used to interpret model outputs and highlight patterns of anomalous behavior. The results validate the model’s applicability for detecting a wide range of cyber threats, including slow-paced attacks, insider threats, and data injections within ETL processes. This paper concludes that AI-driven techniques, particularly those leveraging temporal and contextual data, offer powerful capabilities to secure ETL systems beyond the limitations of traditional methods. Future research will focus on integrating reinforcement learning for dynamic policy updates, real-time deployment in production pipelines, and using federated learning for decentralized data environments. This approach promises not only enhanced security but also improved operational resilience and regulatory compliance.

Spatiotemporal Anomaly-Aware Air Quality Forecasting In South Korea Using Multi-Channel Attention-Based Deep Learning

Juyoung Chang — 2025-04-19

Air quality is becoming a global issue in present days and the monitoring of the air quality is also becoming an important subject for prediction and awareness about air pollutants. In this study an investigation of air quality forecasting has been done with the help of deep learning methods as isolation forest and autoencoders. Data has been collected as sequential data from Korean government meteorological websites from 2018 to 2022 and a spatiotemporal anomaly-aware forecasting is done with graphical attention network combined with LSTM in the encoder part. The study is an integration of spatial correlations among multiple stations of South Korea and the temporal trend and prediction of the pollutants and handling the missing data or outliers in the pollutant reading. Moreover, the incorporation of novel anomaly-aware loss penalizes the outliers more cautiously leads to a stable reading. Experimental results and prediction plots confirm that the proposed model achieves more stable and accurate forecasts. This research highlights the effectiveness of graph-based learning and anomaly-aware strategies in environmental time-series prediction tasks.

Secure and Decentralized Algerian E-Voting System Based on Blockchain and NFC

Hanane Echchaoui — 2025-04-19

The democratic process relies on credible and fair elections for decision-making, which are crucial within communities and in modern democratic countries like Algeria. In recent years, electronic voting systems have gained interest for their potential to minimize costs and enhance efficiency and participation. However, widespread adoption is hindered by security and reliability concerns. As smart cities become more prevalent, it is vital to integrate robust security measures within governmental systems to maintain public trust and safety. Incorporating new, reliable technologies such as blockchain, smart contracts, and NFC into the voting process has the potential to make it quicker, more effective, and less susceptible to security vulnerabilities. This paper introduces an innovative electronic voting system leveraging NFC and blockchain technologies to address these challenges. Our approach aims to ensure security, legitimacy, and trust while employing the Algerian biometric and electronic identity card, which has remained unexploited and underutilized since its launch. Preliminary results highlight the system's potential to revolutionize electronic voting, paving the way for more reliable and secure electoral processes.

A Hybrid Machine Learning and Metaheuristic Framework for Optimizing Time and Cost in Hospital Construction Projects

Reza Zandi Doulabi — 2025-04-19

The rapid aging and functional deterioration of Iran's hospital infrastructure—where over 60% of the 1,100 existing hospitals with 160,000 beds are considered obsolete—pose a critical challenge to achieving national healthcare goals. Moreover, bridging the gap to meet the target of 2.3 hospital beds per 1,000 people requires the addition of approximately 40,000 new beds, amid serious fiscal constraints. This study presents a data-driven decision-support framework to optimize construction time and cost in hospital projects, using actual data from 270 existing facilities. The proposed methodology integrates machine learning models—specifically MLP, SVR, and Random Forest—for predictive analysis, with metaheuristic algorithms including Grey Wolf Optimizer (GWO), Genetic Algorithm (GA), and Artificial Bee Colony (ABC) for multi-objective optimization. Among the predictive models, SVR achieved the highest accuracy in estimating both cost and duration. Optimization results indicated that GWO outperformed the other algorithms, achieving the lowest normalized objective value. In the most efficient scenario, a 108-bed hospital at an optimal location minimized both cost (596 billion Rials) and time (4.45 years), while a fixed-capacity scenario of 300 beds increased both metrics but offered higher service output. The results provide a scalable, evidence-based tool for policymakers and infrastructure planners to evaluate trade-offs between time, cost, and capacity. The approach is particularly useful for strategic healthcare planning under limited resources.

Integrating Blockchain and AI for Data Encryption and Secure ETL Pipelines

Manohar Reddy Sokkula — 2025-04-19

In the era of data-driven decision-making, ensuring the security, transparency, and integrity of Extract, Transform, and Load (ETL) pipelines has become increasingly critical, especially in regulated industries such as healthcare, finance, and telecommunications. Traditional ETL systems often rely on centralized architectures with basic encryption and access control mechanisms, which, although essential, fall short of addressing sophisticated cyber threats, data tampering, and compliance verification. This research proposes a hybrid framework that integrates Blockchain technology and an MLP-GRU (Multi-Layer Perceptron – Gated Recurrent Unit) neural network to enhance the security and intelligence of ETL processes. Blockchain is employed to create a decentralized, tamper-proof ledger that logs each ETL operation, providing traceability, immutability, and auditability. In parallel, the MLP-GRU model is utilized to detect anomalies in ETL activities by analyzing both static and sequential log data. This dual approach ensures not only secure data management but also real-time monitoring and predictive threat mitigation. The experimental setup involves blockchain-based logging of ETL operations and AI-based anomaly detection, evaluated using metrics such as Accuracy (99%), Precision (98.21%), Recall (98%), and F1-Score (98.77%). Results demonstrate that the integrated system outperforms traditional ETL security mechanisms in detecting malicious activity while maintaining efficient data throughput and low latency. Furthermore, the study examines blockchain transaction performance under varying data volumes to validate the scalability of the proposed solution. The framework's ability to automate compliance verification and generate immutable audit trails presents a significant advancement in secure data pipeline design. Future work includes enhancing privacy through Zero-Knowledge Proofs, scaling to federated systems, and incorporating advanced deep-learning architectures. Overall, this research sets a strong foundation for the development of intelligent, secure, and regulation-compliant ETL infrastructures through the convergence of blockchain and AI technologies.

Multilabel Classification for Predicting Crop Pests in Niger

Mahaman Lawali Inoussa Garba — 2025-04-19

Crop pests pose serious threats to agricultural production and food security. With the advent of climate change in Niger, pest attacks have become increasingly frequent. This has become a crucial problem and a priority for farmers and government, as it can destroy the crop or harvest, thereby causing economic harm to the detriment of farmers and the population. Machine learning techniques are widely used in crop pests’ prediction. However, the existing approaches generally focus on the prediction of crop pests using traditional classification methods. These approaches are limited, as they do not make it possible to predict multiple crop pests. Thus, simultaneous and rapid prediction of multiple pests remains a major challenge. In this study, we proposed an approach to predict all the pests of a crop in various localities by using multilabel classification techniques. We developed and compared nine (9) multilabel classification models over two different periods (monthly and annual) using historical data on crop pest infestation and climate. The classifiers are evaluated using Hamming Loss (HL). It was observed that the Radom k-labELsets (RAkEL) classifier is better both on monthly and annual prediction of all pests, with a comparative HL percentage value of 3.63% and 5.1%, respectively. This study extends the models available for crop pest prediction and opens a new path to improving the prediction of crop pests.

Trinetra- A Vision Therapy Application to Aid People with Low Vision

Pushpa G — 2025-04-19

In a more focused analysis, the recent papers stemming from these studies reveal deeper insights into their respective areas. The study cantered on AI in adaptive education uncovers a growing inclination towards neural networks as a potential solution for identifying learning styles. However, despite this interest, there remains a noticeable void in research efforts directed towards comparing and implementing deep learning techniques, highlighting an area ripe for further exploration. In a more focused analysis, the recent papers stemming from these studies reveal deeper insights into their respective areas. The study centred on AI in adaptive education uncovers a growing inclination towards neural networks as a potential solution for identifying learning styles. However, despite this interest, there remains a noticeable void in research efforts directed towards comparing and implementing deep learning techniques, highlighting an area ripe for further exploration. In a more focused analysis, the recent papers stemming from these studies reveal deeper insights into their respective areas.

The Dark Side of AI: How Criminals Leverage Machine Learning for Illicit Activities in the Context of Assault

Kadapa Chenchi Reddy — 2025-04-19

The rise of artificial intelligence (AI) and machine learning has revolutionized various sectors, but it has also opened avenues for malicious use by criminals, particularly in the context of assault. This article explores the dark side of AI, focusing on how criminals leverage these technologies to carry out both physical and cyber assaults. From weaponizing AI-driven drones for targeted attacks to using machine learning for cyberstalking, harassment, and social engineering, criminals are finding increasingly sophisticated methods to exploit these technologies.The article examines real-world examples of AI-assisted assault, including physical and cyber harassment, and the challenges law enforcement faces in detecting and prosecuting such crimes. Additionally, it discusses the ethical and legal implications of regulating AI to prevent its misuse, highlighting the need for stronger safeguards and collaboration between tech companies, policymakers, and law enforcement. As AI continues to evolve, it is essential to balance innovation with ethical responsibility, ensuring that its potential is harnessed for good while mitigating risks to individuals' safety and privacy. The article calls for increased awareness, regulation, and vigilance to safeguard society from the malicious use of AI in criminal activities.

Talk Smart, Talk Small: Crafting Domain-Specific LLMs for SME Customer Support

Inn Keat Ng — 2025-04-19

This project addresses key challenges faced by commercial large language models (LLMs) in customer engagement, such as inconsistent responses, inaccuracies, hallucinations, and lack of follow-up questions. The goal was to develop a domain-specific LLM from scratch for small and medium enterprises (SMEs), capable of delivering relevant, consistent, and human-like responses. The methodology involved studying LLM architectures, preparing and expanding datasets, developing a base model, fine-tuning with larger domain-specific data, applying reinforcement learning, and evaluating model performance. The initial model, trained on 1.5 million tokens, lacked the language understanding needed for coherence. Scaling the dataset to 445 million tokens with general and domain-specific data improved training dynamics and model stability. Fine-tuning with 550 million tokens enhanced relevance, consistency, and human-likeness, outperforming parameter-efficient methods such as LoRA. Reinforcement learning using Identity Preference Optimization (IPO) yielded mixed results. The Normal IPO approach maintained training stability and preserved response quality at both sentence and response levels. However, the Checkpoint and EMA strategies showed fluctuating training behavior and declines in response-level consistency, human-likeness, and relevance, likely due to the small reinforcement learning dataset and instability from evolving reference models. Despite these challenges, the project demonstrated the feasibility of building a domain-specific LLM tailored for SME customer engagement. Future directions include expanding the reinforcement learning dataset, exploring alternative optimization strategies, and incorporating human feedback to further refine performance.

Strategic Design with AI Agents, Predefined Workflows and Agentic AI

Chandrababu C Nallapareddy — 2025-04-19

Artificial Intelligence (AI) agents are transforming application development by enabling adaptive decision-making and dynamic interactions in complex environments. These agents can learn from data, respond to real-time inputs, and operate autonomously, making them powerful for scenarios requiring flexibility and contextual intelligence. However, AI agents are not always the optimal choice. In many cases, predefined workflows structured, rule-based processes deliver greater predictability, cost-efficiency, and maintainability. Recent advances have introduced Agentic AI, a paradigm that blends the adaptability of AI agents with long-term planning, persistent memory, and tool integration. This paper compares AI agents, predefined workflows, and Agentic AI, analyzing their respective strengths, trade-offs, and ideal use cases. It also explores hybrid architectures that combine these approaches, providing practical guidance for selecting the most effective solution based on context, complexity, and operational goals.

Fine-Tuning InceptionV3 for Thai Cuisine Image Classification: A Mobile Deployment Perspective

Somsak Saksat — 2025-04-19

This work presents the development of a smartphone application that utilizes deep learning techniques for the automatic classification of Thai food images. Transfer learning and fine-tuning approaches were compared using the InceptionV3 model, initially trained on the ImageNet dataset and subsequently refined with a dataset consisting of 49 varieties of Thai cuisine images. Experimental results indicate that the fine-tuning model achieved superior performance, attaining an accuracy of 95.22% on the validation set, surpassing the transfer learning model, which achieved an accuracy of 85.43%. Additionally, the fine-tuning model exhibited a stable and consistent decrease in loss without significant overfitting, making it the preferred choice for application development. We converted this model to TensorFlow Lite to enable offline functionality on smartphones developed using Flutter. However, retrieving detailed nutritional information still requires an online database connection to ensure comprehensive nutrient data, including calories, protein, fat, and carbohydrates. This research demonstrates the potential of combining fine-tuning methods with mobile application development to promote mindful food consumption, reduce the risk of non-communicable diseases, and enhance quality of life in the digital era. Moreover, the application supports the United Nations Sustainable Development Goal 3: Good Health and Well-being by encouraging healthier lifestyle choices and contributing to improved health outcomes. Furthermore, it provides a valuable framework for the sustainable promotion of Thai food culture.

Artificial Intelligence: An Analytical Study: Its Impact on Marketing Rate Through Electronic Applications in Saudi Arabia

Sabah Abdellatif Hassan Ahmed — 2025-04-19

This study examines the role of artificial intelligence (AI) in shaping e-commerce trends and its impact on consumer behavior in Saudi Arabia. With the increasing integration of AI technologies, such as machine learning algorithms, chatbots, and virtual assistants, e-commerce platforms are experiencing a transformation in customer engagement, sales performance, and personalized shopping experiences. The research utilizes a mixed-methods approach, combining quantitative and qualitative data from a sample of 349 participants across various regions in Saudi Arabia. The findings reveal that AI-driven features significantly influence online shopping behaviors, with users reporting enhanced satisfaction, increased shopping frequency, and improved decision-making processes through personalized recommendations and dynamic pricing. Key insights from the study include gender-based preferences for specific e-commerce platforms and distinct challenges related to AI adoption, such as privacy concerns, mistrust in AI recommendations, and system complexity. The study also highlights the need for e-commerce platforms to address these challenges by enhancing transparency, refining AI-driven tools, and ensuring a balance between personalization and consumer privacy. Despite these challenges, the data suggest optimism regarding AI's potential to further transform the e-commerce landscape in Saudi Arabia, offering valuable opportunities for businesses to improve customer experiences and operational efficiency. This research provides a comprehensive analysis of AI's impact on digital shopping in Saudi Arabia, contributing to the growing body of knowledge on AI’s influence in the global e-commerce market..

Predicting Hospital Readmission for Diabetes Patients

Hajar Hussein AL Qahtani — 2025-04-19

Predicting hospital readmission among diabetes patients is essential for improving patient outcomes, reducing healthcare costs, and optimizing the use of medical resources. However, this task is complex due to the intricate nature of healthcare data, high feature dimensionality, class imbalance issues, and the necessity of integrating both demographic and clinical variables. To address these challenges, a variety of machine learning models were developed and assessed, including traditional classifiers such as Decision Trees, Logistic Regression, and Random Forests, as well as more advanced approaches like XGBoost and Deep Neural Networks. To enhance model performance, we applied preprocessing techniques such as feature transformation, data balancing, and categorical encoding. Experiments were conducted on clinical datasets to predict patient readmission within 30 days, after 30 days, or not at all. Performance metrics included classification accuracy and the AUC-ROC score. Results showed that the Random Forest model achieved the highest performance in binary classification, with an accuracy of 94% and an AUC-ROC of 0.97, while a proposed Multi-Stage Classifier excelled in the multi-class task with 80% accuracy and an AUC-ROC of 0.89. Overall, the study highlights the potential of machine learning, particularly when coupled with effective preprocessing, to accurately predict hospital readmissions in diabetes care, thereby aiding clinical decisions and improving healthcare efficiency.

Strengthening Banking Security: Pioneering AI-Driven Identity and Access Management Solutions

Avinash Chowdary Vikram — 2025-04-19

The banking industry is currently grappling with complex challenges in Identity and Access Management (IAM), which are essential for safeguarding sensitive customer information and maintaining regulatory compliance. As financial institutions evolve and adapt to an increasingly digital landscape, robust security measures are more crucial than ever to protect against a rising tide of sophisticated cyber threats. Traditional IAM approaches often fall short in addressing these evolving threats, as they may lack the agility and intelligence needed to respond effectively. This underscores the urgent need for AI-driven solutions that not only enhance security but also improve operational efficiency.This study delves into the transformative role of Artificial Intelligence (AI) in revolutionizing IAM within the banking sector. By focusing on critical components such as AI-driven authentication, fraud prevention, and risk-based access control, the study illustrates how innovative technologies can mitigate risks while ensuring compliance with regulatory frameworks. For instance, AI enhances authentication processes by utilizing advanced algorithms that analyze user behavior and patterns, making it significantly more difficult for unauthorized users to gain access.The paper includes real-world case studies that demonstrate the effectiveness of AI-based IAM solutions in enhancing security protocols, providing insights into successful implementations that have resulted in reduced instances of fraud and improved customer trust. Additionally, these case studies showcase how financial organizations have leveraged AI to create a more streamlined access control system, allowing users to navigate financial services seamlessly while maintaining a stringent security posture.

A Hybrid Model for Detection of Breast Cancer through Efficient Feature Selection using Machine Learning Approaches

Pradip Chakraborty — 2025-04-19

Cancer in breasts is considered as one of the dreaded diseases. It causes huge loss of human lives throughout the world and its menace is spreading fast. Earlier detection of breast cancer significantly enhances treatment effectiveness and patient’s prognosis. Traditional methods in many cases of diagnosis incur much expenses, time taking and prone to errors resulting demoralization and unsuccessful. Machine learning approaches have been showing promises in automating detection of cancers in breasts. There exist a number of approaches in machine learning which show good results. This research tries to find out the techniques from the existing models and by addressing and modifying the underlying technical issues towards attaining higher accuracy. This study works using three individual learners namely, ‘Support Vector Machines, ‘Logistic Regression’ and ’Decision Trees’. Derives a hybrid learner from these three individual leaners .Using a comprehensive dataset obtained from clinical studies, available publicly online the proposed model applies Principal Component Analysis (PCA) for selecting features. This approach processes dataset, discern subtle patterns and enhances diagnostic accuracy reducing human errors. In the comparative analysis, it presents results of the model and evaluation of performance through metrics like accuracy, sensitivity and specificity. The model finally achieves 98.24% of accuracy in prediction which seems to be impressive in comparison to other existing models. The study upholds its potential as a significant tool in medical diagnostics.

Data-Driven Tax Management: Developing a Long-Run Analysis System for Niger through Outlier Detection

Moussa Khane — 2025-09-17

Developing countries often face challenges in conducting long-term economic analyses, which in turn affects their ability to design effective planning and policy decisions. This study applies the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework to support Niger’s tax administration in implementing a long-run analysis scheme. Several statistical and machine learning tools, such as boxplots, the interquartile range (IQR), the augmented Dickey-Fuller test, the Johansen test, and the vector error correction model (VECM) were employed. The dataset covers the period from January 1996 to December 2014 and reveals seven (7) outliers. Results showed that VAT, ITS, and ISB account for 93.64% of revenues in the dataset with outliers and 94.25% without outliers, confirming that cointegration tests were highly sensitive to outliers. Both datasets were non-stationary but cointegrated, with rank three (3). Tax revenue took approximately 81 days to absorb shocks with outliers, compared to 60 days without. Outliers, thus, significantly distorted the Nigerien economic planning and policy outcomes.