Fake News Detection Utilizing Ensemble Learning and Explainable AI

The rapid spread of misinformation poses a significant threat to public trust and decision-making. This research involves training XGBoost and LightGBM classifiers on the ISOT and WELFake datasets, employing TF-IDF, BERT, and Word2Vec for feature extraction, followed by SHapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) for explainability analysis of the model predictions. Among all configurations, the best performance was achieved by combining Word2Vec with LightGBM, reaching 0.979 accuracy and 0.980 F1-score on the ISOT dataset. SHAP effectively highlighted globally important features from sparse TF-IDF inputs, while LIME provided localized insights for dense embeddings. These results demonstrate the potential of explainable ensemble methods for fake news detection across varying text styles, offering a scalable and transparent solution, with future research in multilingual and multi-modal contexts.