Enhancing the Predictive Performance of Credibility-Based Fake News Detection Using Ensemble Learning

Abstract

Fake news detection continues to be a major problem that affects our society today. Fake news can be classified using a variety of methods. Predicting and detecting fake news has proven to be challenging even for machine learning algorithms. This research employs Legitimacy, a unique ensemble machine learning model to accomplish the task of Credibility-Based Fake News Detection. The Legitimacy ensemble combines the learning potential of a Two-Class Boosted Decision Tree and a Two-Class Neural Network. The ensemble technique follows a pseudo-mixture-of-experts methodology. For the gating model, an instance of Two-Class Logistic Regression is implemented. This study validates Legitimacy using a standard dataset with features relating to the credibility of news publishers to predict fake news. These features are analysed using the ensemble algorithm. The results of these experiments are examined using four evaluation methodologies. The analysis of the results reveals positive performance with the use of the ensemble ML method with an accuracy of 96.9%. This ensemble’s performance is compared with the performance of the two base machine learning models of the ensemble. The performance of the ensemble surpasses that of the two base models. The performance of Legitimacy is also analysed as the size of the dataset increases to demonstrate its scalability. Hence, based on our selected dataset, the Legitimacy ensemble model has proven to be most appropriate for Credibility-Based Fake News Detection.

Research of news text classification method based on hierarchical semantics and prior correction

“…For example, there are text classification methods based on the K-nearest neighbors algorithm such as [7,8] and based on support vector machines including [9][10][11]. In recent years, many papers have used a variety of machine learning algorithms for news text classification to explore the most appropriate approach for a particular domain [12][13][14][15]. Among them, Sidar [12] uses 41 diverse machine learning algorithms to classify 4000 news articles in specific domains, successfully showcasing the potential of various machine learning algorithms.…”