SugarCheck Insights: A Diabetes Prediction and Risk Profiling System Using Machine Learning

Abstract

SugarCheck Insights is a full stack web application which can be used to predict diabetes and risk profile. The system combines a custom implementation of a Support Vector Machine (SVM) based binary classification and the K-means based unsupervised stratification of patients build using three phased incremental methodology. The project is developed based on a balanced dataset of 17,000 records, out of 100,000 original records, and further preprocessed and validated the models. In order to reduce false negatives, it is ensured that the custom SVM is optimized over algorithms such as Logistic Regression and Naive Bayes giving it a better accuracy of 88.56% and a high recall of 90.59%. K-means clustering is effective in the grouping of patients as low, medium and high risks. The last unified system offers user authentication, role-based dashboards, and real-time prediction and reflects a viable, end-to-end system of proactive management of diabetes.