Predictive Modeling for Arabic Fake News Detection: Leveraging Language Model Embeddings and Stacked Ensemble

Abstract

The proliferation of fake news poses a substantial threat to information integrity, prompting the need for robust detection mechanisms. This study advances the research on Arabic fake news detection and overcomes the limitation of lower accuracy for fake news detection. This research addresses Arabic fake news detection using word embedding and a powerful stacking classifier. The proposed model combines bagging, boosting, and baseline classifiers, harnessing the strengths of each to create a robust ensemble. Extensive experiments are carried out to evaluate the proposed approach indicating remarkable results, with recall, F1 score, accuracy, and precision reaching 99%. The utilization of advanced stacking techniques, coupled with appropriate textual feature extraction, empowers the model to effectively detect Arabic fake news. Study results make a valuable contribution to fake news detection, particularly in the Arabic context, providing a valuable tool for enhancing information veracity and fostering a more informed public discourse. Furthermore, the proposed model’s accuracy is compared with other cutting-edge models from the existing literature to showcase its superior performance.