Fake News Detection Using TF-IDF Feature Extraction and Supervised Machine Learning Classifiers

The rapid spread of fabricated news articles on social media platforms poses a serious threat to public opinion, democratic processes and social harmony. Manual fact checking cannot keep pace with the volume of content generated every day, which motivates the development of automated detection systems. This paper presents the design, implementation and evaluation of a machine learning based fake news detection system that classifies English news articles as real or fake using stylometric and lexical cues. Articles from the ISOT fake news corpus are preprocessed through lowercasing, punctuation removal and stop word elimination, and are then transformed into Term Frequency Inverse Document Frequency vectors comprising five thousand unigram and bigram features with sublinear term frequency scaling. Three supervised classifiers, namely Logistic Regression, Linear Support Vector Machine and Random Forest, are trained and compared on an eighty twenty stratified split. The Linear Support Vector Machine achieves the best performance with 99.77 percent accuracy, 99.85 percent precision and 99.72 percent recall on 8854 unseen test articles, misclassifying only 20 samples. The trained model is deployed as an interactive Streamlit web application that returns a verdict, a confidence score and a word level explanation derived from the product of model coefficients and feature weights. The results confirm that carefully engineered sparse lexical features with linear classifiers remain a strong, computationally inexpensive baseline for fake news detection.