Hybrid System for Diabetes Prediction

Abstract

Diabetes is a chronic condition that can be caused by the body’s inability to produce or use insulin effectively. Over time, this can result in damage to various organs, including the heart, blood vessels, eyes, kidneys, and nerves. The timely detection of diabetes is essential for its prompt treatment, as it can halt the progression of the disease. In this study, we propose an hybrid machine learning approach to predict diabetes using a com- bination of two powerful algorithms. We used an ensemble learning based on Deep Neural Network and Random Forest classifier, and Support Vector Machine as a meta classifier (SVC). We trained and tested our model on the Pima Indian diabetes dataset , which contains 77568 instances and 8 features, using 5-fold cross validation. Our experimental results show that our proposed approach achieved an accuracy of 95% , outperforming other state of the art machine learning techniques. We propose also an alternative ap- proach that combines random forest and XGBoost by a voting technique getting 92.7% of accuracy. Our findings suggest that the hybrid machine learning approachs we proposed can be used as a reliable tool for early diabetes detection, enabling more timely and ef- fective interventions to improve patient outcomes.