Performance Analysis of Diabetes Forecasting Using Machine Learning Techniques

This paper analyzes how three different machine learning models can predict diabetes evaluated using a health record dataset that includes BMI, glucose levels, and age of patients. Multiple algorithms were studied: Logistic Regression Model, Decision Trees, Gradient Boosting, Support Vector Machines (SVM) and Random Forest. The best performance was observed in Gradient Boosting as it achieved an accuracy of 96.8%, a precision of 98.5%, recall of 98.5% and an F1 score of 99.2%, thus good at predicting. Random Forest also achieved fair results, at 94.6% accuracy, therefore deemed to be a good alternative. Logistic Regression and Decision Trees had much lower accuracies, 80.4% and 86.8% respectively. Feature importance analysis revealed glucose concentration, BMI and age to be the most relevant factors consistent with the existing medical literature. The results demonstrate the validity of using multi-models to enhance early detection and management of diabetes more so Gradient Boosting. Forthcoming works aim at incorporating these models in real health monitoring and in prediction of customer behavior.