Diabetes Prediction using Machine Learning

Diabetes mellitus is a long-term metabolic disorder, which sometimes causes serious complications in the heart, kidneys, nerves, and eyes when not detected in time. In light of the effect of the high-risk factors like obesity, age, insulin resistance, hypertension and lifestyle imbalance, intelligent diagnostic support syste5s are necessary to identify risks early. This paper will suggest a Pima Indians Diabetes Dataset-based and machine learning-based smart diabetes prediction model. Preprocessing of data was done to deal with missing and invalid values and to deal with class imbalance by making use of SMOTE and ADASYN. GR, SVM, KNN, RF, GB, and XGBoost were selected as multiple classifiers and trained and optimized with the help of GridSearchCV and tested on the basis of the accuracy, precision, recall, F1-score, and ROC-AUC measures. It was experimentally found that the ADASYN with XGBoost had the highest performance with an accuracy of 82.47% and an AUC of 0.90. Explainable AI methods including SHAP and LIME were used to improve clinical interpretability through risk factors identification. The results show that the proposed machine learning-based diagnostic support can be very helpful in detecting diabetes at an earlier stage and aid in healthcare decision-making.