Comparison of ECG Between Gameplay and Seated Rest: Machine Learning-Based Classification

The influence of gameplay on autonomic nervous system activity was investigated by comparing electrocardiogram (ECG) data during seated rest and gameplay. A total of 13 participants (6 in the gameplay group and 7 in the control group) were analyzed. RR interval time series (2 Hz) and heart rate variability (HRV) indices, including mean RR, SDRR, VLF, LF, HF, LF/HF, and HF peak frequency, were extracted from ECG signals over 5-minute and 10-minute segments. HRV indices were calculated using fast Fourier transform (FFT). The classification was performed using Logistic Regression (LGR), Random Forest (RF), XGBoost (XGB), One-Class SVM (OCS), Isolation Forest (ILF), and Local Outlier Factor (LOF). A balanced dataset of 5-minute and 10-minute segments was evaluated using k-fold cross-validation (k = 3, 4, 5). Performance metrics, including recall, F-score, and PR-AUC, were computed for each classifier. Grid search was applied to optimize parameters for LGR, RF, and XGB, while default settings were used for the other classifiers. Among all models, OCS with k = 3 achieved the highest classification accuracy for both 5-minute and 10-minute data. These findings suggest that machine learning-based classification can effectively distinguish ECG patterns between gameplay and rest.