Arabic Fake News Detection Model Using Ensemble Machine Learning Classifiers

Abstract

Information sharing on social media has grown faster, less costly and easily accessible. This useful information may contain fake news in many fields including politics, medicine and sports for various reasons such as advertising and propaganda. Ability to identify, analyze and address such information is significantly important. Many studies have been conducted to detect fake news in English, but there is a lack in Arabic language. This paper exhibits model to detect Arabic fake news on X-Platform during Ukraine-Russia conflict with the assistance of Ensemble Machine learning. In this work, an Arabic fake news dataset was collected related to the Ukraine-Russia conflict. Six different classification algorithms are trained to classify news as fake or real and are compared considering accuracy, recall, precision. The experimental results demonstrated that Random Forests (RF) using Term Frequency-Inverse Document Frequency (TF-IDF) preforming with SMOTE techniques achieved the best predictions with an accuracy of 99 % and testing accuracy with 98 %. Indeed, the results show that applying ensemble algorithms with TF-IDF and SMOTE achieved better improvement on evaluation metrics compared to the baseline classifier and other classifiers without SMOTE.