Deep Learning vs Compression-Based vs Traditional Machine Learning Classifiers to Detect Hadith Authenticity

Due to the increasing numbers of Hadith forgeries, it has become necessary to use artificial intelligence to assist those looking for authentic Hadiths. This paper presents detailed research on ways to automatically detect Hadith authenticity in Arabic Hadith texts. It examines the utilization of deep learning-based and prediction by partial matching (PPM) compression-based classifiers, which have not been previously used in detecting Hadith authenticity. The proposed methods were compared with the most recent method used which is machine learning. In addition, there is a detailed description of the new Arabic Hadith corpus (non-authentic Hadith corpus) created for this study and the authors’ experiments, which also used the Leeds University and King Saud University (LK) Hadith corpus. The experiments demonstrate that the authentication based on Isnad obtained accuracy ranging from 84% to 93%. The authentication based on Matan obtained an accuracy range of 55% to 93%, while the accuracy range for this experiment was from 55% to 85%, which means that Isnad is the most effective part of Hadith for automatically detecting authenticity. Moreover, the experiment proved that Matan can be used to judge Hadith authenticity with an accuracy of 85%. The study also showed that PPM and deep learning classifiers are effective means of automatically detecting authentic Hadith.