A refined SMOTE-ENN optimization method based on machine learning for heart rate variability data classification.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Computational Neuroscience

Background:

• Imbalanced heart rate variability (HRV) data presents challenges for machine learning models in depression detection.
• Early identification of depression is crucial and can be supported by analyzing Autonomic Nervous System (ANS) states.

Purpose of the Study:

• To propose a refined SMOTE-ENN hybrid optimization method for precise classification of imbalanced HRV data.
• To enhance machine learning algorithm performance for early depression detection using HRV analysis.

Main Methods:

• A refined Synthetic Minority Over-sampling Technique (SMOTE) and Edited Nearest Neighbor (ENN) under-sampling algorithm were developed.
• Four machine learning algorithms (SVM, Random Forest, Neural Network, KNN) were applied to optimized HRV data from 321 participants.

Main Results:

• All four machine learning algorithms achieved over 91% classification accuracy and AUC values exceeding 0.92 after refined SMOTE-ENN optimization.
• Compared to classical SMOTE, the refined method showed significant improvements in accuracy, precision, recall, and F1 score.
• Standard deviation of NN intervals (SDNN) was identified as the most influential feature in HRV classification.

Conclusions:

• The refined SMOTE-ENN method effectively enhances machine learning performance for imbalanced HRV data classification.
• This approach offers valuable technical support for the early detection of depression through improved ANS state analysis.