Comparative analysis of machine learning methods to detect fake news in an Urdu language corpus

Wide availability and large use of social media enable easy and rapid dissemination of news. The extensive spread of engineered news with intentionally false information has been observed over the past few years. Consequently, fake news detection has emerged as an important research area. Fake news detection in the Urdu language spoken by more than 230 million people has not been investigated very well. This study analyzes the use and efficacy of various machine learning classifiers along with a deep learning model to detect fake news in the Urdu language. Logistic regression, support vector machine, random forest (RF), naive Bayes, gradient boosting, and passive aggression have been utilized to this end. The influence of term frequency-inverse document frequency and BoW features has also been investigated. For experiments, a manually collected dataset that contains 900 news articles was used. Results suggest that RF performs better and achieves the highest accuracy of 0.92 for Urdu fake news with BoW features. In comparison with machine learning models, neural networks models long short term memory, and multi-layer perceptron are used. Machine learning models tend to show better performance than deep learning models.