Comparative Study of Machine Learning Approaches for Detecting Fake News in Arabic Text

Abstract

It is evident that fake news remains a critical global problem, especially in the Arabic language, although there is an absence of vast amounts of annotated datasets required for effective stateof- the-art natural treatment. In this paper, we compare deep neural networks (DNNs), XGBoost, gradient boosting (GB), and long short-term memory (LSTM) networks on the task of distinguishing real and fake Arabic news. When we applied special preprocessing for AFND with specific approaches to tackle the class imbalance problem, we observed that XGBoost was found to be the best method, performing with an accuracy of 72.86% on the test database. The present model performed optimally on relevant parameters related to the absence of capitalized terms, precision (0.83%), recall (0.71%), and F1-score (0.76%), especially for “undecided” cases. XGBoost’s performance is revealed in these results, and feature selection and optimization are promoted, leading to improvements in the Arabic natural language processing (NLP) domain.