Efficient Machine Learning Approach for Diabetes Mellitus Disease Prediction

Diabetes mellitus, a widely recognized chronic disease often termed a silent killer, disrupts insulin production, leading to elevated blood sugar levels and complications affecting organs including the eyes, kidneys, and nerves. Despite significant research interest, diabetes classification-based machine learning methods suffer from low performance due to missing data and imbalanced class data. Thus, this paper introduces a new method for Diabetes Mellitus classification for Imbalanced data with Missing values (DMIM). The proposed approach involves these key steps. Firstly, missing values are preprocessed using the Isolation Forest method that provides data normalization. Then, an adaptive synthetic sampling technique, Smote T-link, is employed to counter the impact of class imbalance. Finally, we utilized a Random Forest (RF) classifier and conducted experiments on the Pima Indians diabetes dataset from the University of California. Our proposed approach yielded highly accurate predictions, as evidenced by impressive performance metrics such as an Area Under the Curve (AUC) of 97.2%, Specificity (SP) of 91.7%, and Recall (Rec) of 92.3%, Further comparison between other ML methods including k-nearest Neighbors, Decision Trees, Random Forest, AdaBoost, Naive Bayes, XGBoost, Extra Trees. The results demonstrate the effectiveness and superiority of our proposed method and improved diabetes classification performance compared to existing techniques.