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

The classification of imbalanced heart rate variability (HRV) data utilizing machine learning algorithms is of considerable significance for the early detection of depression. In this work, a refined SMOTE-ENN hybrid optimization method based on machine learning algorithms is proposed, which achieves precise classification of Autonomic Nervous System (ANS) states using imbalanced and limited HRV data.

The refined Synthetic Minority Over-sampling Technique generates new minority class samples through in-line and off-line linear interpolation. Subsequently, the refined Edited Nearest Neighbor under-sampling algorithm is employed to remove most of the noisy data while retaining selected boundary data to reduce the overfitting risk. Four machine learning algorithms are employed to classify the optimized HRV data using the refined SMOTE-ENN method from 321 participants, including support vector machine, random forest, neural network, and K-nearest neighbors.

The results indicate that the classification accuracy of all four methods surpasses 91%, with the AUC (Area Under the Curve) values exceeding 0.92 following the refined SMOTE-ENN optimization. In comparison to the classification results with classical SMOTE optimization of the four machine learning algorithms, the mean accuracy, precision, recall, and F1 score improved by 0.12, 0.12, 0.10 and 0.11, respectively. Feature importance analysis reveals that SDNN (standard deviation of NN intervals) has the most significant impact on HRV classification results, reflecting its influence on ANS.

The refined SMOTE-ENN optimization method enhances the detection performance of the machine learning algorithms for classifying imbalanced HRV data, providing valuable technical support for the early detection of depression.