Enhancing Diabetes Detection using Machine Learning: A Focus on Optimizing Recall Performance

Diabetes mellitus is a clinically and genetically heterogeneous group of disorders that can be distinguished by high blood glucose levels caused by either insulin deficiency or the body's resilience to the action of insulin. Diabetes detection at an early stage is critical for preventing damage to internal organs and the immune system. Most of the applications focus on optimizing the model based on accuracy rather than recall; a good recall score is required because it tells us how accurately the relevant class is being predicted, and a wrong prediction would lead to undiagnosed treatment. This paper proposes the most optimal prediction model to diagnose diabetes. The dataset from BRFSS includes a person's health indicators, which possess key features for us to analyze the effectiveness of different machine learning algorithms and observe the best performing model.