Improving Prediction of Arabic Fake News Using ELMO’s Features-Based Tri-Ensemble Model and LIME XAI

The proliferation of fake news poses a substantial and persistent threat to information integrity, necessitating the development of robust detection mechanisms. In response to this challenge, this research specifically focuses on the detection of Arabic fake news, employing a sophisticated approach that leverages textual features and a powerful stacking classifier. The proposed model ingeniously combines bagging, boosting, and baseline classifiers, strategically harnessing the unique strengths of each to create a resilient ensemble. Through a series of extensive experiments and the integration of Embeddings from Language Models (ELMO) word embedding, the proposed approach achieves remarkable results in the realm of Arabic fake news detection. The model’s effectiveness is further heightened by the utilization of advanced stacking techniques, coupled with meticulous textual feature extraction. This capability enables the model to effectively distinguish between real and fake news in Arabic, highlighting its potential to enhance the accuracy of information. The findings of this study hold significant implications for the field of fake news detection, especially within the context of the Arabic language. The proposed model emerges as a valuable tool, contributing to the enhancement of information veracity and fostering a more informed public discourse in the face of misinformation challenges. Furthermore, the excellence of the proposed model is substantiated by its outstanding performance metrics, boasting a 99% accuracy, precision, recall, and F-score. This substantiation is underscored through a comprehensive performance comparison with other state-of-the-art models, affirming the model’s reliability in the domain of Arabic fake news detection.