Fake News Prediction and Analysis in LIAR Dataset Using Advanced Machine Learning Techniques

Abstract

Fake news detection and prediction is the crucial research issue in now a day because it is very difficult to know the news authenticity on social media. It has a devastating impact on societies and democratic institutions as online life in these days are one of the principal news hotspots for many individuals around the world because of their minimal effort, simple access, and quickly spread of the unauthorized news. However, measurable ways to deal with battling fake news have been drastically restricted by the absence of named benchmark datasets. Smart machine learning classifiers are used to solve the problem of fake news prediction and classification. The proposed research study works on the LIAR dataset, the open-source available dataset for fake news classification with 12.8K decade-long, hand-labelled short statements in various contexts. The proposed research study has used a novel approach to deal with the fake news prediction accurately and this approach outperforms in this scenario for the same dataset. Naïve Bayes classifier for classification is used to reduce the variance values in the dataset to get rid of the overfitting issue. This classifier shows more improved results than other prior classifiers and the accuracy value was 99%. The proposed research study performed experiments and evaluated through different evaluation measures, the results of accuracy for the Naïve Bayes are best as compared to Random Forest, Decision tree, and Neural Networks are computed for each algorithm. The proposed research study could be applied in real-time applications to deal the fake news prediction in social media and digital media platforms.