Early Prediction of Diabetes Mellitus: An Explainable AI Approach

Abstract

Objectives: This study aims to propose a model to predict early Diabetes Mellitus (DM) using Explainable Artificial Intelligence (XAI) to complement clinical decisions while maintaining both high accuracy and explainability. Methods: The general model to be developed in this study involves applying Ensemble Machine Learning algorithms with interpretability plans. Crossvalidation is applied to construct a model that will be further used to select the most significant risk factors from a public database containing patient data. To make sure its results can be interpreted by humans, the model employs rule extraction and feature importance analysis. The model’s accuracy and efficiency to diagnose are evaluated by using a variety of datasets. Findings: In predicting DM, the XAI model secured a sensitivity of the work at 92% and specificity at 88%, offering accurate and interpretable results. Feature importance analysis and rule extraction were helpful to clinicians to improve understanding and trust in the model because the output of both methods results in a form that is human consumable. Novelty: This work fills the existing gap in black-box machine learning models by increasing model interpretability and applying state-of-art techniques to ensemble methods. This indicates that the proposed framework achieves both high accuracy and high interpretability, letting to DM diagnosis and efficient disease control from the beginning. That is why this study presents a new direction in developing new models that fill the gap between high performance and interpretability and provide valuable insights and results that are meaningful from the clinical perspective. Keywords: Diabetes Mellitus (DM); Explainable Artificial Intelligence; Ensemble Machine Learning; LIME; Predictive Modelling; SHAP