Fake News Classification Using Logistic Regression with TF-IDF Feature Extraction

Abstract

Abstract. The rapid proliferation of fake news on digital platforms poses a significant threat to public knowledge, democratic processes, and social cohesion. Existing detection methods often lack interpretability or require substantial computational resources, creating a research gap for lightweight yet accurate approaches. This study presents a machine learning-based fake news detection system employing Logistic Regression combined with Term Frequency-Inverse Document Frequency (TF-IDF) vectorization. The WELFake dataset, comprising 72,134 labeled news articles, was utilized after preprocessing steps including lowercasing, URL removal, digit elimination, punctuation stripping, and whitespace normalization. Row containing missing value were removed prior to splitting, yielding 71,537 records. The dataset was then split 80:20 for training and testing with stratified sampling, resulting in a test set of 14,308 samples. The proposed model achieved an accuracy of 95.0%, precision of 94.15%, recall of 96.17%, F1-score of 95.15%, and an AUC of 0.9884, demonstrating strong discriminative capability. The novelty of this work lies in the systematic integration of TF-IDF feature engineering with Logistic Regression for the WELFake benchmark, accompanied by comprehensive exploratory analysis including text-length profiling, word clouds, and coefficient-based feature importance interpretation. Results confirm that interpretable linear models remain highly competitive against more complex architectures for text classification tasks in the misinformation domain.

 

Keywords: fake news detection, logistic regression, machine learning , TF-IDF, WELFake dataset