AI Based Fake News Detection Using NLP

The proliferation of social media and digital news platforms has made the rapid, uncontrolled spread of fabricated and misleading content — commonly termed fake news — a pressing societal challenge, capable of manipulating public opinion, influencing elections, and eroding trust in legitimate journalism. Manual fact-checking, while accurate, is inherently too slow to match the pace of digital content distribution. This paper presents an AI-based Fake News Detection system that applies Natural Language Processing (NLP) and Machine Learning (ML) to classify news articles as Real or Fake in real time. Raw text is normalized through a fixed pre-processing pipeline — cleaning, lowercasing, tokenization, stop-word removal, and stemming/lemmatization — before being converted into numerical feature representations using Bag-of-Words, TF-IDF, Word2Vec/GloVe embeddings, and BERT-based contextual embeddings. These features are passed to a suite of interchangeable classifiers, including Naive Bayes, Support Vector Machine, Logistic Regression, Passive-Aggressive, Random Forest, LSTM networks, and transformer-based BERT models, allowing an explicit trade-off between accuracy and computational cost. A coefficient-driven explainability layer highlights the specific keywords and sentences that most influenced a given verdict, and all predictions are logged to a relational database for auditing and analytics. The system is delivered as a full-stack web application — a React front end and a FastAPI back end — and was validated through a layered testing strategy spanning unit, integration, functional, system, white-box, black-box and acceptance testing, with fifteen functional test cases achieving a 100% pass rate. Benchmarking of the classical classifiers indicates that Naive Bayes achieves an accuracy of 88%, an F1-score of 88.3%, and a precision of 85%, confirming that a computationally lightweight model remains competitive while transformer-based models offer superior contextual accuracy at higher computational cost. The results confirm that a scalable, explainable, and freely deployable fake-news screening tool can materially reduce the gap between the speed of misinformation and the speed of verification.