Implementation And Testing of Machine Learning Framework for Fake News Detection

Abstract

The rapid spread of fake news across online platforms threatens public trust and information integrity. This paper presents an advanced machine learning framework for fake news detection using two benchmark datasets: the LIAR dataset and the Kaggle Fake/Real News dataset. This proposed approach combines classical models such as Logistic Regression with advanced models including LightGBM and embedding-based classifiers. Further, incorporation of explainability techniques such as LIME and SHAP has been done for predictions and enhancement in transparency. Experimental results demonstrate that LightGBM achieves superior performance, while cross-dataset evaluation reveals moderate generalization capability. The proposed system provides both high accuracy and interpretability, making it suitable for smart information systems.