EARLY STAGE DIABETES RISK PREDICTION AND PREVENTION SYSTEM USING ML

Abstract

Diabetes mellitus is a chronic metabolic disorder that affects a very large and growing number of people, and a considerable share of them remain undiagnosed until complications such as nerve damage, kidney disease, or vision loss have already begun. Early detection allows lifestyle changes and medical treatment to slow or even prevent the progression of the disease. This paper presents an early stage diabetes risk prediction and prevention system that uses machine learning to estimate an individual's risk from simple symptoms, clinical measurements, and lifestyle information, and then provides practical prevention guidance based on that risk. The system uses two kinds of input. The first is a symptom questionnaire covering signs that commonly appear in the early stage of diabetes, such as excessive urination, excessive thirst, sudden weight loss, weakness, blurred vision, delayed healing, itching, and irritability, together with age and gender. The second is a set of clinical and lifestyle measurements, including fasting glucose, body mass index, blood pressure, waist circumference, family history, physical activity, and dietary habits. Where available, readings from a glucometer, a digital blood pressure monitor, and a weighing scale can be entered directly, which reduces transcription errors. The collected data is cleaned, encoded, and scaled, and missing clinical values are handled carefully because zero readings in such datasets often represent missing measurements rather than true values. Several classifiers are trained and compared, including logistic regression, k-nearest neighbours, a support vector machine, a decision tree, random forest, and gradient boosting. Because missing a diabetic case is more harmful than a false alarm, the models are evaluated with particular attention to recall and F1 score in addition to accuracy, precision, and the area under the ROC curve. The selected model assigns each user to a low, moderate, or high risk category along with a probability value. The prevention module then generates advice that matches the risk level and the specific factors contributing to it, such as increasing daily physical activity, reducing refined carbohydrate intake, managing weight, or consulting a physician for a confirmatory HbA1c test. A web dashboard allows users to record repeated measurements over time and view their risk trend, while health workers can review the risk profile of a screened community. The system is intended as a screening and awareness tool rather than a diagnostic instrument, and every high-risk result is accompanied by a recommendation to seek laboratory confirmation. It can be used in college health centres, primary health camps, and workplaces where large numbers of people can be screened quickly at low cost. Future enhancements include integration with Bluetooth-enabled glucometers and wearable activity trackers, support for regional languages, and periodic retraining with locally collected data.