Comprehensive Analysis of Machine Learning and Deep Learning Models for Fake News Detection on Twitter

Abstract

The rapid growth of online platforms has resulted in an increase in the spread of fake news, causing significant social harm. This study examines machine learning and deep learning methods for detecting fake news across a number of datasets. Logistic Regression, Random Forest, LSTM, Bi-LSTM, and a Continuous Attention Mechanism Embedded Bi-LSTM (CAME) were evaluated for their performance on a Twitter dataset. The exponential rise of internet platforms has caused an upsurge in the propagation of fake news, causing immense social harm. This study examines machine learning and deep learning methods for detecting fake news across a number of datasets, achieving accuracy levels up to 79%. Logistic Regression produced an 81% accuracy with precisions of 0.80 with class 0 and 0.82 of class 1. Random Forest obtained 79% accuracy with precise of 0.77 with class 0 as 0.82 of class 1. The LSTM model achieved 76.43% accuracy with peak training accuracy of 91.33%, Bi-LSTM achieved 77.68% accuracy with peak training accuracy of 91.73%, and the CAME model achieved 79.05% accuracy with peak training accuracy of 91.95%.