An Exploration of Deep Learning Algorithm for Fraud Detection using Spark Platform

Abstract

Fraudulent activities pose an important threat in several areas, requiring robust and efficient mechanisms for detection. It is critical to halt fraudulent transactions since they have a long-term influence on financial circumstances. Anomaly detection has several essential applications for detecting fraud. This paper presents a novel fraud detection method using deep learning algorithms, combining Convolutional Neural Networks (CNNs) for feature extraction from transaction data and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies in financial transactions, thereby enhancing robust and efficient detection mechanisms. This paper proposes a new framework that combines Spark with a deep learning technique. However, it compares the performance of deep learning approaches for credit card fraud detection with other machine learning algorithms, including CNN-LSTM, on three distinct financial datasets. This paper also employs several machine learning algorithms for fraud detection, such as random forest, SVM, and KNN. Various parameters are used in comparative analysis. Both the training and testing datasets achieved more than 96% accuracy. The text outlines the creation of a high-performance deep learning model for detecting credit card fraud. The paper proposes hybrid attention to integrate current time output with unit state, determining its weight, and optimizes accuracy using Adam optimization. It uses various machine learning methods for a comparative study using the proposed deep architecture.