Background and Aims: Diabetes mellitus, a group of metabolic diseases requiring multifactorial risk reduction and continuous medical care, poses significant challenges for complications prevention beyond glycemic control. This study addresses the contemporary emphasis on Artificial Intelligence (AI) and machine learning (ML) to develop algorithms capable of learning patterns and decision rules from data. Despite the existence of risk scores for complications, their limitations in accurately estimating both types of diabetes complications underscore the need for predictive models based on local data applicable in bedside and clinic settings.

Methods: In the Endocrinology OPD setting of a tertiary care hospital, we have utilized four Neural Network-based algorithms (Random Forest, Decision Tree, K Neighbour [KNN], and Artificial Neural Networks [ANN]) to predict complications in type 2 diabetes patients.

Results: A Deep Neural Network model integrating these algorithms achieved optimal results, particularly with the ANN GRU model exhibiting a sensitivity of 89%, specificity of 97%, F1 score of 0.93, and AUC ROC of 0.98.

Conclusion: This study outlines the successful development and validation of a machine learning-based model for predicting adverse outcomes associated with diverse diabetes complications, underscoring the potential of machine learning in individual risk predictions and offering a practical application for patient education, facilitating behavior change for risk reduction and overall wellness.