DiabetCare – A Classifier based Mobile Application for Predicting the Risk of Prediabetes Mellitus, Gestational Diabetes Mellitus and Type 2 Diabetes Mellitus

Diabetes mellitus, commonly referred to as diabetes, poses a significant global public health challenge. According to the International Diabetes Federation, the number of people affected by this condition is projected to rise from 415 million to 642 million by 2040. This chronic ailment disrupts the body's ability to absorb glucose, making early risk prediction essential for diagnosis and prevention. This research aims to introduce an innovative AI-driven method for early risk assessment of three distinct types of diabetes. Utilizing medical data from the University of Kelaniya, two different ensemble learning techniques are implemented. These predictions will be accessible through an automated mobile application named "DiabetCare," enabling risk evaluation irrespective of age. The objective is to identify and manage diabetes in its early stages, thereby reducing potential future complications. The experimentation process encompassed the training of 12 machine learning base classifiers and three ensemble techniques for each component. Optimal parameters for each algorithm were determined using the random search cv hyper-parameter tuning method. Ultimately, the most effective base and ensemble models, based on evaluation metrics, were chosen for integration into the final product. In terms of predicting the risk of type 2 diabetes or prediabetes, the Random Forest classifier achieved perfect training accuracy (1.0000), a testing accuracy of 0.8715, an AUC-ROC score of 0.9811, and a Cohen's Kappa of 0.8037. For forecasting the risk of gestational diabetes, the Stacking classifier attained a perfect training accuracy (1.0000), a testing accuracy of 0.9650, an AUC-ROC score of 0.9963, and a Cohen's Kappa of 0.9288. The implementation of such a system holds immense potential within the field of diabetes medicine and can serve as a valuable validation tool for medical experts.