Author(s) :

Rudra Mogre.

Volume/Issue :

Abstract :

The proliferation of fake news across online platforms poses a growing challenge to information integrity and public trust. Traditional fact-checking mechanisms are often too slow to counteract the viral spread of misinformation. As a result, automated approaches using machine learning have emerged as effective tools for distinguishing between legitimate and fabricated news articles. This study applies supervised machine learning techniques to classify news content using the ISOT (~44,898 articles) and LIAR (12,836 statements) benchmark datasets. Feature extraction was performed using Term Frequency- Inverse Document Frequency (TF-IDF) and n-gram analysis (unigrams and bigrams). Five classical algorithms were evaluated: Logistic Regression, Naïve Bayes, Support Vector Machines (SVM), Decision Trees, and Random Forests. Model performance was assessed using accuracy, precision, recall, F1- score, and AUC. On the ISOT dataset, Random Forest achieved the highest accuracy of 0.998, followed by SVM and Decision Tree at 0.996, Logistic Regression at 0.993, and Naïve Bayes at 0.949. On the LIAR dataset, accuracies ranged from 0.571 (Decision Tree) to 0.628 (Logistic Regression). The results demonstrate that classical machine learning models, coupled with robust text representation techniques, can effectively detect misinformation on structured full-article datasets like ISOT while maintaining transparency and scalability, though performance drops significantly on short-claim datasets like LIAR due to structural differences.