A clinically-informed approach to predictive modelling of Diabetes Mellitus for the East and West Godavari Districts

The ever-increasing prevalence of Diabetes Mellitus (DM) in India, particularly in diverse regional populations, demands the development of highly accurate, localized predictive models for early diagnosis and intervention. This study focuses on the Godavari districts of Andhra Pradesh, a region with distinct lifestyle and dietary patterns that influence diabetes risk. To develop and validate a state-of-the-art Artificial Neural Network (ANN) model for the tri-state classification of individuals into ‘Healthy’, ‘Pre-Diabetes’, and ‘Diabetes’. The model utilizes a comprehensive dataset of clinical, lifestyle, and continuous glucose monitoring (CGM) parameters from this specific regional cohort to achieve exceptionally high diagnostic accuracy. We utilized a cross-sectional, balanced dataset of 2,617 individuals from the East and West Godavari districts. A multi-layer perceptron ANN was designed and trained on 22 features, including demographic data, anthropometric measurements, glycemic markers (Fasting Glucose, Postprandial Glucose, HbA1c), and lifestyle factors. The model’s performance was meticulously evaluated on a held-out test set of 655 samples, using a confusion matrix to derive accuracy, precision, recall, and F1-score. The optimized ANN model demonstrated exceptional diagnostic performance, achieving a near-perfect overall classification accuracy of 99% on the test set. The model distinguished between the three classes with remarkable precision and recall. It perfectly identified all diabetic individuals (Recall: 1.00) and achieved F1-scores of 0.99 for the ‘Healthy’ class and 0.98 for the ‘Pre-Diabetes’ class, indicating an extremely low rate of misclassification. Our findings establish the superior efficacy of an ANN-based approach for diabetes risk stratification in the Godavari region. The model’s outstanding accuracy in classifying individuals into distinct glycemic states emphasizes its potential as a highly reliable clinical decision support tool. This level of precision can significantly enhance early detection, enabling timely and targeted interventions to prevent the progression of diabetes and mitigate its public health burden.