Ensemble-Based Machine Learning Approach for Detecting Arabic Fake News on Twitter

Abstract

The rise of social media platforms has led to a significant increase in the spread of false or misleading information, which has become a major issue of concern. Twitter faces the difficult task of identifying and reducing the spread of 'fake news', which refers to material that is erroneously or intentionally spread. This type of content frequently includes false information, biased information, and data that is provided without considering its full context. The swift and comprehensive proliferation of platforms such as Twitter worsens the problem by enabling the widespread and rapid dissemination of unverified content, often leading to its viral dissemination and contributing to the propagation of falsehoods. This paper presents a specialized machine learning approach that utilizes an ensemble-based strategy to identify and classify false information on the social media platform Twitter. This approach utilizes the combined power of various classifiers, such as Support Vector Machines (SVM), k-Nearest Neighbors (KNN), and Gradient Boosting (GBoost), to create a strong prediction model by combining multiple weaker learners. Every classifier undergoes rigorous training using a specific set of variables, including text content, user profile information, and tweet metadata. This enables a thorough analysis to identify fake news. Within the proposed system, following separate training, the classifiers generate predictions that are then merged. A neural network is utilized to combine the outputs of all classifiers, resulting in a definitive prediction. This approach tackles the drawbacks of overfitting and improves the capacity of the model to apply to new data, resulting in a higher level of accuracy for the machine learning model. The empirical assessments conducted on a dataset that is freely accessible, containing both genuine and counterfeit tweets, show that the ensemble model performs much better than the individual base classifiers and traditional machine learning models. The proposed method attained an accuracy of 0.963, along with an Area Under the Curve (AUC) of 0.964, surpassing the precision, recall, and F1 scores of its individual classifiers. The results confirm the efficacy of the ensemble machine learning architecture as a dependable method for identifying false information in Arabic on Twitter. This has implications for wider usage on different social media platforms.