A comparative approach to alleviating the prevalence of diabetes mellitus using machine learning

Diabetes mellitus, a metabolic disease with elevated blood sugar levels, is a significant global public health concern. Identification of diabetes at its very early stage can reduce the prevalence of cases. This work focuses on developing a machine learning-based system that will have a significant impact on diabetic patient identification. To develop such a system we utilized a dataset made up by acquiring direct questionnaires from Sylhet Diabetic Hospital patients. The dataset contains information about the signs and symptoms of patients who are new or likely to have diabetes. We applied 14 different machine-learning techniques where the Gradient Boosting Machine (GBM) outperformed other algorithms with the highest F1 and ROC scores of 99.37%, and 99.92% respectively. We also employed various ensemble-based approaches that show competitive performance to the individual model’s performance.