Utilizing Machine Learning to Aid in Type 2 Diabetes Diagnosis

High blood glucose levels are a defining characteristic of diabetes, with type 2 diabetes being the most prevalent form. Early detection and treatment can prevent or delay complications. To precisely anticipate type 2 diabetes risk, we applied machine learning techniques to identify unidentified correlations with a person's health condition. In order to find potential nonlinear correlations between the beginning of type 2 diabetes with an ensemble of health factors, this research used a binary classifier that was trained from scratch. Three datasets—MIMIC-III, MIMIC-IV, and NHANES—were used. The equal number of people who had or did not have type 2 diabetes have been included in one data set that was created by combining the two sets of data. Reliability was the key metric, given the balanced dataset. The model achieved ROC AUC score of 0.934 with accuracy ratings up to 86%, which were positive results of this investigation. To increase the model's dependability, more research is required to take into account multiple readings taken throughout period for an identical person.