A Symbolic AI Framework for Enhanced Diabetes Prognosis Accuracy and Explainability

Le résumé fourni par la source

Machine learning offers a promising avenue for accurate disease prediction, facilitating early detection and prevention. However, traditional models often grapple with challenges such as the opacity of their decision-making processes and the substantial resources required to improve model explainability. To address these issues, this paper introduces a novel symbolic AI-based framework that synergistically integrates domainspecific knowledge graphs (KGs) with machine learning models to enhance efficiency and explainability. Knowledge graphs provide a structured representation of domain-specific knowledge, making AI models more transparent and interpretable. By incorporating KG embeddings into various machine learning models, we aim to identify the optimal combination for accurate prediction while elucidating the model's decision-making process. We applied our methodology to the PIMA Indian diabetes dataset, experimenting to validate our approach. Given that diabetes poses a significant global health challenge, necessitating practical prognostic tools for timely intervention, our methodology enhances predictive accuracy. It identifies substantial features contributing to the prediction of type 2 diabetes. Using KGs for AI model explainability presents a novel and efficient method, reducing the time and effort compared to traditional approaches. Our findings demonstrate that symbolic AI approaches, utilizing KGs as embeddings, significantly outperform conventional AI models, providing a robust and interpretable solution for diabetes prognosis. Data and source code available on Github.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.