Fake News Detection Using Machine Learning: A Comparative Analysis of Classification Algorithms on Fake News Net Dataset

Abstract

This paper presents a comprehensive empirical study on fake news detection using machine learning techniques applied to the FakeNewsNet dataset comprising 23,196 news articles. We systematically compare three classification algorithms: Logistic Regression, Random Forest, and XGBoost, employing TF-IDF text vectorization with n-gram features and SHAP explainability analysis. Our methodology incorporates sophisticated preprocessing and feature engineering, achieving robust performance with 84.25% accuracy and 90.14% F1score. Feature importance analysis through SHAP reveals that sensational language patterns (”breaking”, ”secret”, ”exposed”) strongly indicate fake news, while formal reporting language (”according to”, ”official”, ”report”) characterizes real news. The research demonstrates that ensemble methods combined with explainable AI techniques provide both high accuracy and interpretable insights into fake news detection. Our findings offer practical deployment strategies and highlight the importance of transparent AI systems in combating misinformation.