IMPROVING HEALTHCARE PREDICTION OF DIABETIC PATIENTS USING KNN IMPUTED FEATURES AND TRI-ENSEMBLE MODEL

COMPUTER SCIENCE AND ENGINEERING

Diabetes Mellitus (DM) is a prevalent chronic disease that poses significant health and economic burdens worldwide. Early detection and accurate prediction of diabetes are crucial for timely intervention and improved patient outcomes. However, real-world medical datasets often contain missing values, which can adversely impact the performance of machine learning models. This study proposes an automated diabetes prediction framework that integrates K-Nearest Neighbors (KNN) imputation for handling missing data and a novel Tri-Ensemble Model for enhanced classification accuracy. The proposed Tri-Ensemble Model combines Extreme Gradient Boosting (XGB), Random Forest (RF), and Extra Trees Classifier (ETC) using a voting mechanism to make robust predictions. The dataset, sourced from the Pima Indians Diabetes Database, was preprocessed to address missing values using KNN imputation, and the models were trained using a 70:30 train-test split. The experimental results demonstrate that the proposed Tri-Ensemble Model significantly outperforms traditional machine learning classifiers. With an accuracy of 97.49%, precision of 98.16%, recall of 99.35%, and F1-score of 98.84%, the model surpasses existing state-of-the-art approaches. Additionally, a comparative analysis reveals that imputing missing values using KNN substantially improves model performance compared to deleting missing data. The findings of this research highlight the importance of effective data preprocessing and ensemble learning techniques in medical diagnostics. The proposed model holds promise for real-world healthcare applications, facilitating early diabetes detection and improving patient care. Future research will explore deep learning models to further enhance predictive accuracy in diabetes diagnosis.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.