Application of irregular and unbalanced data to predict diabetic nephropathy using visualization and feature...

Area of Science:

• Nephrology
• Data Science
• Machine Learning

Background:

• Diabetic nephropathy is a common diabetes complication, leading to kidney damage.
• Predicting its onset is challenging due to patient variability and unidentified determinants.
• Traditional statistical methods lack accuracy for early prediction.

Purpose of the Study:

• To accurately predict the onset of diabetic nephropathy using machine learning techniques.
• To apply support vector machine (SVM) classification and feature selection to irregular, unbalanced diabetes datasets.
• To develop a visualization system for intuitive risk factor analysis.

Main Methods:

• Collected medical data from 292 diabetic patients, extracting 184 features.
• Compared logistic regression, SVM, and cost-sensitive SVM for prediction.
• Utilized feature selection methods to enhance classification performance.
• Developed a nomogram-based visualization system for SVM risk factor analysis.

Main Results:

• Linear SVM with wrapper or embedded feature selection achieved the highest performance.
• Classifiers identified 39 key features out of 184.
• Achieved an area under the curve of 0.969 via receiver operating characteristics analysis.
• The visualization tool effectively presented feature impact graphically.

Conclusions:

• The proposed machine learning method predicts diabetic nephropathy onset 2-3 months earlier than traditional methods.
• Achieved high prediction performance on complex, irregular, and unbalanced datasets.
• The visualization system provides physicians with intuitive risk factor insights.
• Facilitates early intervention and personalized treatment strategies for diabetic nephropathy.