Improving Healthcare Prediction of Diabetic Patients Using KNN Imputed Features and Tri-Ensemble Model

Objective: Diabetes ranks as the most prevalent ailment in developing nations. Vital steps to mitigate the consequences of diabetes include early detection and expert medical intervention. A highly effective approach for identifying diabetes involves assessing the specific indicators associated with this condition. When it comes to automated diabetes detection, frequently encountered datasets frequently exhibit gaps in data, which can markedly impact the effectiveness of machine learning models. Methods: The aim of this study is to propose an automated method for predicting diabetes, with a focus on appropriately dealing with missing data and improving accuracy. The proposed framework makes use of K-Nearest Neighbour (KNN) imputed features along with a Tri-ensemble voting classifier model. Results: By incorporating the KNN imputer, the presented model demonstrates impressive performance metrics, including an accuracy of 97.49%, precision of 98.16%, recall of 99.35%, and an F1 score of 98.84%. The study conducted a thorough comparison of this proposed model against seven alternative machine learning algorithms, assessing them under two conditions: one with omitted missing values and another with the KNN imputer applied. These findings support the proposed model’s efficacy, highlighting its superiority over currently established state-of-the-art techniques. Conclusion: This research explores the problem of missing data in diabetes diagnosis and highlights the efficacy of the KNN-imputed technique. The results are promising for healthcare practitioners as they could facilitate early detection and improve the quality of diabetic patient care.