Interpretable Machine Learning Approach for Real-Time Diabetes Risk Prediction

Abstract

Diabetes mellitus is one of the most prevalent chronic metabolic diseases and represents a major global public-health challenge in the 21st century. It is characterized by persistently elevated blood glucose levels caused by insufficient insulin production, insulin resistance, or both mechanisms. Type 2 Diabetes Mellitus (T2DM) accounts for most cases and is strongly associated with lifestyle factors such as obesity, sedentary behaviour, unhealthy dietary habits, and genetic predisposition [1]. Long-term uncontrolled diabetes can lead to severe complications including cardiovascular diseases, neuropathy, retinopathy, kidney failure, and reduced life expectancy [2].

Traditional diagnostic methods such as fasting plasma glucose tests, oral glucose tolerance tests, and glycated haemoglobin (HbA1c) assessments are clinically reliable but often require laboratory infrastructure and repeated testing [3]. With the rapid growth of healthcare datasets and electronic health records, machine learning techniques have emerged as effective tools for early disease prediction and healthcare analytics [7], [8].

This paper proposes a machine learning-based diabetes prediction system using a Decision Tree classifier. The model analyses several clinical parameters including glucose level, body mass index (BMI), lipid profile indicators, kidney function markers, and age-related attributes to predict diabetes risk. Data preprocessing techniques such as normalization, missing value handling, and class imbalance correction are applied to improve model performance. The trained model is integrated into a Flask-based web application capable of classifying individuals into non-diabetic, pre- diabetic, and diabetic categories. The system demonstrates the potential of interpretable machine learning models for early risk detection and decision support in modern healthcare systems.