A Predictive Fake News Detection Model Using Machine Learning

There is an enormous increase in the amount of fake news being distributed via digital platforms, which greatly endangers public trust and societal stability. Classical machine-learning-based text classification techniques, while truly effective, often face challenges in differentiating between an equally convincing piece of misinformation and a completely valid news piece. This study attempts to propose a hybrid fake news detection framework that encompasses text-based classification coupled with external verification using web scraping and gradation of credibility for a given piece of news from social media. Machine-learning algorithms including Logistic Regression, Random Forest, and Naïve Bayes were trained on real and fake news articles in a labelled dataset, achieving a best accuracy of 99.7 %. In an effort to strengthen the classification's credibility, a web scraper was designed to check the validity of news articles against trusted media sources using fuzzy matching techniques and achieved an increase in trust score of real news from 0.45 to 0.73. Besides, another trust scoring mechanism was developed on social media user credibility referring to engagement pattern, network reliability, and online presence of the sources of news. Ultimately the integration of all the identification approaches yielded an overall classification accuracy of 96.1 %, exhibiting an appreciable improvement over traditional text-based systems. The results signify the need for multi-pronged approaches to fake news detection combining automated text classification with external credibility assessments. Future work will look into optimizing the speed of web scraping to handle a larger dataset in parallel with some real-time monitoring of the system to maintain robustness to newer tricks of misinformation.