Detection of Depression-Specific Circadian Misalignment in Young People via Whole-Day HRV and Actigraphy Patterns

Circadian misalignment has been increasingly associated with depressive symptoms in young individuals, contributing to the persistence and recurrence of depression. Timely detection of such misalignment is therefore critical for early intervention. This study investigates circadian rhythm disturbances in physiological responses linked to depression and proposes a computational framework for symptom visualization and automated detection. A total of 47 features were extracted from electrocardiogram, photoplethysmography and tri-axial accelerometer signals collected over 1084 days from 148 participants. Daily recordings were divided into nonoverlapping time slots, resulting in a 47 × 153 feature matrix per individual per day. We introduced a novel index, the parasympathetic activation index (PAI), to quantify circadian variation in parasympathetic activity and to capture depression-related disruptions. Statistical analyses indicate that individuals with depressive symptoms exhibit significant circadian misalignment, characterized by elevated nighttime PAI and reduced daytime PAI. A deep neural network model trained on this feature representation achieved an F1-score of 85% under a leave-one-subject-out cross-test scheme. These results highlight the feasibility of leveraging wearable physiological signals and machine learning for real-world, user-independent depression monitoring and early risk identification in young populations.