A Computational Investigation of Fabricated News Identification Using TF-IDF and Machine Learning

Abstract

The rapid expansion of digital communication has made it increasingly difficult to distinguish reliable journalism from fabricated or deliberately misleading information. False stories, altered claims, and unsupported statements can circulate within seconds through websites, social-media platforms, and online discussion forums. Manually checking every article is time-consuming and difficult to scale. To address this challenge, this study develops a text-based fake-news detection system that combines Natural Language Processing, Term Frequency–Inverse Document Frequency (TF-IDF), and supervised Machine Learning. The news dataset was cleaned, normalised, and converted into TF-IDF feature vectors before model training. Logistic Regression, Multinomial Naive Bayes, and Random Forest were trained and evaluated under the same experimental conditions. After preprocessing, 38,605 valid records remained, and the TF-IDF representation was restricted to 5,000 important features. Model performance was measured using accuracy, precision, recall, F1-score, and confusion matrices. Random Forest achieved the highest accuracy of 99.62%, followed by Logistic Regression at 98.54% and Multinomial Naive Bayes at 93.46%. These findings demonstrate that traditional machine-learning methods can effectively identify patterns associated with fabricated news within the selected textual dataset. However, the results remain dependent on the dataset, preprocessing strategy, feature configuration, and model settings used in the experiment.