Sleepiness detection in daily life based on multimodal signals

This study employs machine learning methods with the aim of detecting daily-life sleepiness in the daily study life of college students, facilitating individuals to adjust their states sensibly and enhance work efficiency. Initially, we collected electrocardiogram (ECG) and tri-axial accelerometer data from 83 college students during their daily activities. We extracted 21 features from the ECG and 3 features representing the level of physical activity from the tri-axial accelerometer data. Subsequently, we trained five classifiers using these features and employed a backward selection algorithm through Leave-One-Subject-Out (LOSO) cross-validation to choose the most crucial feature subset. Ultimately, utilizing the best-performing classifier and the selected key feature subset, we constructed a model for detecting daily-life sleepiness, achieving the F1 score of 77.61% on the independent test set. The results suggest that employing machine learning to detect the sleepiness of individuals is feasible.