FAKE NEWS DETECTION USING MACHINE LEARNING ALGORITHMS

The rapid proliferation of digital communication networks, online journalism 
platforms, and user-generated social media content has fundamentally transformed modern 
information dissemination. However, this shift has simultaneously enabled the unvetted 
propagation of intentionally fabricated, misleading, and biased reporting—commonly 
conceptualized as fake news. The sheer volume and velocity of digital content render manual 
content moderation by human domain experts entirely unscalable. Automated, intelligent detection 
mechanisms leveraging Natural Language Processing (NLP), Data Science techniques, and 
Machine Learning (ML) classifiers have consequently emerged as critical components of digital 
information defense infrastructure. In this research paper, we design, implement, and rigorously 
evaluate an end-to-end framework for computational fake news identification. The architecture 
incorporates comprehensive raw text normalization, noise reduction, tokenization, stop-word 
elimination, and Porter Stemming, followed by term-weight representation using Term Frequency
Inverse Document Frequency (TF-IDF) n-gram vectorization. Six prominent supervised machine 
learning classifiers—Logistic Regression, Support Vector Classifier (SVC), Decision Trees, 
Random Forests, Multinomial Naive Bayes, and Passive Aggressive Classifiers—are benchmarked 
on extensive real-world corpora. Experimental outcomes demonstrate that linear online 
algorithms, specifically the Passive Aggressive Classifier and Logistic Regression, yield superior 
classification performance, achieving validation accuracies of 94.25% and 93.80% respectively.