AN INTERPRETABLE MODEL FOR AMHARIC HADITH TEXTS AUTHENTICITY CLASSIFICATION: A DL APPROACH

Abstract

The Prophetic Hadith constitutes a pivotal foundation of Islamic jurisprudence, second only to the Quran. Serving as a compass for Muslims seeking guidance, Hadith encapsulates the words, deeds, and tacit consent of the Prophet Muhammad (P.B.U.H). Yet, the proliferation of non authentic Hadiths has sown seeds of doubt and uncertainty, especially among Amharic speaking Ethiopian Muslims, who face challenges in authenticating these narratives due to a lack of computational tools. Addressing this, our study developed an automated system for classifying Amharic Hadith texts utilizing deep learning algorithms and explainable AI techniques. We collected and annotated a dataset of 16,654 Amharic Hadiths from five esteemed canonical books- Sahih al-Bukhari, Sahih Muslim, Sunan Abu Dawud, Jami al Tirmidhi, and Sunan Ibn Maja. The texts were labeled with 12 authenticity grades, each of these grades has a different level of authenticity and reliability according to the Islamic scholars. Employing various deep learning architectures, we pursued an optimal model that not only categorizes Hadiths by authenticity but also elucidates its reasoning via SHapley Additive exPlanations (SHAP). The combined CNN-BiLSTM model emerged superior, boasting an accuracy of 89% and surpassing baseline classifiers. The implementation of SHAP identified influential narrators in the sanad and prevalent themes in the matn driving predictions. This novel system is poised to revolutionize Hadith authentication for Ethiopian Muslims, facilitating access to verified texts and enriching their understanding of the Islamic canon. Our work transcends technical achievement, significantly contributing to computational linguistics for lesser-studied languages and enhancing the resources available to religious communities. Looking ahead, the challenge is to broaden the dataset and enhance the translation accuracy from Arabic to Amharic, aiming to capture the full spectrum of linguistic nuances and thematic richness. The way forward entails a comprehensive expansion of corpus size and diversity, rigorous scholar-led validation in the translation to ensure fidelity to the original texts, further integration of linguistic context, and a continuous dialogue between AI development and ethical standards to ensure inclusivity and cultural sensitivity in serving the global Islamic community