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
