Leveraging Traditional Machine Learning and TF-IDF for Robust Fake News Detection

Social media platforms have created space for fabricated news to circulate broadly that currently has become a leading problem distorts public opinion and disrupts political discourse and credibility of online information. The research analyzes how traditional machine learning (ML) models detect fake news from the identification of text content with engineered language features. The experiment proves that traditional methods provide competitive performance with less computational power and maintaining greater interpretability if augmented by strong preprocessing and feature extraction methods. A comprehensive preprocessing pipeline consisted of text normalization followed by stop word removal stemming and n-gram modeling and TF-IDF vectorization to transform raw text to numerical features. The performance was assessed by stratified cross-validation and held-out test set on three supervised ML algorithms, i.e., Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF). The performance assessment included accuracy along with precision and recall and F1-score and ROC-AUC. The results showed that Logistic Regression provided maximum accuracy (98.1%) and F1-score but Random Forest provided maximum recall rate (98.5%) and therefore it was better at detecting actual fake news. The SVM model provided well-balanced outcomes but it was computationally expensive. Traditional ML models show high effectiveness in detecting fake news with proper feature engineering techniques.