The proliferation of misinformation on social media poses a critical threat to public discourse and societal stability. Existing fake news detection systems primarily focused on high-resource languages like English, often function as "black box" classifiers that lack the interpretability and cross-lingual generalization necessary for real-world trust and deployment, especially in multilingual, low-resource settings. This research addresses this dual challenge by proposing an Explainable Multilingual Fake News Detection Framework built on state-of-the-art Transformer models. We leverage the power of pre-trained models such as mBERT and XLM-R to develop a robust detector for challenging, low-resource Indian languages, specifically Hindi and Marathi, alongside English. The core contribution of this work is the integration of an Explainable AI (XAI) module utilizing techniques such as LIME and SHAP. This module provides crucial transparency by highlighting the specific linguistic and semantic cues (keywords and sentences) that drive the model’s prediction, thereby enhancing user trust and accountability—a critical need frequently overlooked in current research. A comprehensive comparative study will benchmark the performance (Accuracy, F1-score, and Interpretability) of our Transformer-based approach against traditional Machine Learning and Deep Learning models (e.g., SVM, CNN/LSTM), demonstrating significant improvements in robustness and explanatory power. Finally, the framework will be deployed as a prototype web application for real-time fake news detection on social media feeds, moving the research from theory to practical application. This work provides a foundation for more transparent, reliable, and linguistically-inclusive fake news detection systems.

Keywords: Fake News Detection; Machine Learning; Explainable Artificial Intelligence (XAI); Multilingual Models; Transformer Architecture; mBERT; XLM-R; LIME; SHAP; Deep Learning; Cross-Lingual Transfer; Natural Language Processing (NLP); Transparency.