OE-MDL: Optimized Ensemble Machine and Deep Learning for Fake News Detection

Abstract

The escalating spread of fake news in the modern digital landscape has sparked significant concerns over the reliability and integrity of online content. Identifying and mitigating fake news is critical for protecting individuals, organizations, and broader society from the adverse effects of misinformation. However, conventional fake news detection methods, such as rule-based systems, supervised machine learning, and natural language processing (NLP) techniques, are impeded by notable drawbacks. Rule-based strategies are rigid and inflexible, supervised learning models often fail to generalize beyond their training data, and NLP methods struggle to fully understand the subtleties and context of language. In response to these challenges, this study presents the Optimized Ensemble Machine and Deep Learning (OE-MDL) algorithm, a sophisticated approach designed to efficiently and accurately detect fake news. The OE-MDL algorithm enhances detection capabilities by incorporating a series of preprocessing steps: converting text to lowercase, tokenization, eliminating stop words, applying word stemming and lemmatization, and conducting spell-checks. It also involves generating n-grams and calculating term frequency-inverse document frequency (TF-IDF) scores, capturing a wide spectrum of linguistic and statistical features that help distinguish between genuine and fraudulent news. The OE-MDL framework enhances classification precision and robustness by integrating optimized machine learning (OML) and optimized deep learning (ODL) phases. In the OML phase, advanced classifiers, including optimized RandomForest, J48, SMO, LSTM, NaiveBayes, and IBk, are amalgamated with an optimized Multilayer Perceptron serving as the Meta classifier. This amalgamation forms the foundation for a bagging classifier, which is then utilized within an AdaBoostM1 boosting classifier. Similarly, the ODL phase employs a Dl4jMlpClassifier as a basis for another bagging and AdaBoostM1 boosting sequence. The OML and ODL classifiers are then synergized through a blending classifier using weighted voting to accurately categorize the training data. The well-trained blending classifier is subsequently deployed to determine the authenticity of news articles in the test dataset. Empirical results underscore the superior performance of the OE-MDL algorithm, achieving unprecedented accuracy (99.87%), precision (99.88%), recall (95.87%), and F1-Score (99.96%). This performance indicates that the OE-MDL algorithm is an exceptionally effective tool in the ongoing battle against the proliferation of fake news, providing a robust and reliable means of upholding the integrity of information in the digital age.