Diabetes mellitus is a growing public health challenge disproportionately affecting low- and middle-income countries, where rural communities often lack access to laboratory-based diagnostic infrastructure. This study presents a proof-of-concept machine learning model for non-invasive diabetes screening using four readily obtainable features: Body Mass Index (BMI), Age, High Blood Pressure (HighBP), and General Health status (GenHlth). A Random Forest classifier was trained on the CDC Behavioral Risk Factor Surveillance System (BRFSS) 2015 dataset, comprising 70,692 balanced records. The model achieved an overall accuracy of 73%, precision of 71%, and a recall of 77% for diabetic cases. The model was successfully deployed to Google Cloud Vertex AI, demonstrating production-level deployability. This work demonstrates that a simple, low-cost screening tool based on non-invasive inputs can be developed and deployed for use by community health workers in under-resourced rural settings.
Tesfaye H. Kifle (Sat,) studied this question.