An automated approach to predict diabetic patients using KNN imputation and effective data mining techniques

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Diabetes is a prevalent global health issue, particularly in developing nations.
• Early detection and management are critical for mitigating diabetes complications.
• Existing automated diabetes detection methods often suffer from accuracy issues and struggle with missing data.

Purpose of the Study:

• To develop a robust automated diabetes prediction system.
• To address the challenge of missing data in diabetes datasets.
• To enhance the accuracy and reliability of diabetes detection models.

Main Methods:

• A stacked ensemble voting classifier model integrating three machine learning algorithms was employed.
• K-Nearest Neighbors (KNN) Imputer was utilized for effective handling of missing data.
• The proposed model was rigorously compared against seven other machine learning techniques in scenarios with and without data imputation.

Main Results:

• The proposed model achieved exceptional performance metrics: 98.59% accuracy, 99.26% precision, 99.75% recall, 99.45% F1 score, and 99.24% MCC.
• The KNN Imputer significantly improved model performance compared to methods that simply eliminated missing values.
• The developed model demonstrated superior efficacy over existing state-of-the-art methods for diabetes detection.

Conclusions:

• The study highlights the effectiveness of the KNN Imputer in managing missing data for improved diabetes prediction.
• The proposed stacked ensemble model offers a highly accurate and robust solution for automated diabetes detection.
• Findings can aid medical professionals in earlier problem identification and enhanced diabetes patient care.