Comparison of Different Machine Learning Techniques to Predict Diabetic Kidney Disease

Area of Science:

• Medical Informatics
• Computational Biology
• Nephrology

Background:

• Diabetic kidney disease (DKD) is a progressive complication of diabetes, leading to kidney function loss.
• Early identification of high-risk individuals is crucial for timely intervention and improved patient outcomes.
• Machine learning classification offers a potential method for building predictive models to screen at-risk populations.

Purpose of the Study:

• To evaluate and compare the performance of various machine learning classification techniques for predicting DKD.
• To identify the most suitable classification method for early detection of diabetic kidney disease.

Main Methods:

• Analysis of nine distinct classification techniques applied to a DKD dataset (410 instances, 18 attributes).
• Data preprocessing using PartitionMembershipFilter and 10-fold cross-validation.
• Performance assessment based on accuracy, execution time, error rates, kappa statistics, and confusion matrix values.

Main Results:

• IBK and random tree classification techniques demonstrated superior performance with 93.6585% accuracy and a high kappa value (0.8731).
• These models exhibited the lowest root mean squared error (0.2496).
• The best models resulted in 15 false positives and 11 false negatives.

Conclusions:

• IBK and random tree are identified as the most effective classifiers for predicting diabetic kidney disease.
• These techniques provide accurate prediction methods for early DKD detection.