Predicting cardiac autonomic neuropathy category for diabetic data with missing values.

Area of Science:

• Medical Informatics
• Diabetology
• Cardiology

Background:

• Cardiovascular autonomic neuropathy (CAN) is a significant diabetes complication.
• Existing methods for CAN data with missing values involve data deletion or feature imputation.
• These traditional approaches can lead to information loss and reduced classification accuracy.

Purpose of the Study:

• To introduce and evaluate a novel method for classifying CAN data with missing values.
• To compare the new method against traditional approaches that delete data or add features.
• To identify the optimal regression and meta-regression combination for CAN classification.

Main Methods:

• The study proposes a new classification method based on regression and meta-regression techniques.
• This method uniquely integrates the Ewing formula for CAN class identification.
• It avoids deleting attributes with missing values and does not employ traditional classifiers or feature addition.

Main Results:

• The best performing method, an additive regression meta-learner (M5Rules) combined with the Ewing formula, achieved 99.78% accuracy for two CAN classes and 98.98% for three classes.
• These results significantly surpass previous findings in the literature.
• The method demonstrated superior performance without requiring additional data collection.

Conclusions:

• The novel regression-based approach effectively handles missing values in CAN data classification.
• This method offers a more accurate and efficient alternative to traditional data imputation and deletion techniques.
• It simplifies the diagnostic process by eliminating the need for supplementary tests.