Journal of affective disorders · 2025
Actigraphy-based step analysis for the detection of depressed mood: An explainable machine learning approach.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Kim Ju-Wan, Lee Taeyeong, Lim Bahngtaik, Park Se-Hyoun, Park Jin-Hyun, Jeong Inyong, Park Kiwon, Kang Hee-Ju, Jeon Eunkyoung, Kim Sung-Wan, Jhon Min, Lee Hwamin, Kim Jae-Min
Аннотация
INTRODUCTION: Our study aims to develop an interpretable artificial intelligence (AI) model for detecting depressive symptoms using actigraphy, integrating statistically significant features into machine learning models to enhance accuracy and explainability.
METHODS: We analyzed actigraphy data from 3304 participants over a one-week period, classifying them into a depressive symptom group and a non-depressive symptom group. Six machine learning models, including CatBoost (CB) and XGBoost (XGB), were trained using absolute activity indicators based on three-hour intervals, relative activity indicators (nonparametric, Cosine analysis), and daytime light intensity exposure duration. Shapley additive explanations (SHAP)-based explainability analysis was applied, and models were stratified by and gender.
RESULTS: CB and XGB demonstrated the highest classification performance in predicting mild and moderate-to-severe depressive symptoms, respectively, with AUROC values of 0.679 and 0.715 across 10 random-seed models evaluated on a fixed test set. SHAP-based explainability analysis revealed that low step counts contributed to the prediction of depressive symptoms, high average activity during the least active 5-hour period of the day (L5) was associated with depressive symptoms, and we found that the onset of L5 was mainly distributed in the dawn hours. Importantly, model performance differed across demographic groups, with the highest predictive accuracy achieved in older men (AUROC = 0.756 and 0.833, respectively) for all depressive symptoms.
CONCLUSION: This study highlights the potential of actigraphy-derived step count data in AI-driven classification of depressive symptoms. Both absolute step count and temporal patterns contribute to classification, emphasizing the need for time-sensitive, explainable AI approaches for personalized mental health screening.
METHODS: We analyzed actigraphy data from 3304 participants over a one-week period, classifying them into a depressive symptom group and a non-depressive symptom group. Six machine learning models, including CatBoost (CB) and XGBoost (XGB), were trained using absolute activity indicators based on three-hour intervals, relative activity indicators (nonparametric, Cosine analysis), and daytime light intensity exposure duration. Shapley additive explanations (SHAP)-based explainability analysis was applied, and models were stratified by and gender.
RESULTS: CB and XGB demonstrated the highest classification performance in predicting mild and moderate-to-severe depressive symptoms, respectively, with AUROC values of 0.679 and 0.715 across 10 random-seed models evaluated on a fixed test set. SHAP-based explainability analysis revealed that low step counts contributed to the prediction of depressive symptoms, high average activity during the least active 5-hour period of the day (L5) was associated with depressive symptoms, and we found that the onset of L5 was mainly distributed in the dawn hours. Importantly, model performance differed across demographic groups, with the highest predictive accuracy achieved in older men (AUROC = 0.756 and 0.833, respectively) for all depressive symptoms.
CONCLUSION: This study highlights the potential of actigraphy-derived step count data in AI-driven classification of depressive symptoms. Both absolute step count and temporal patterns contribute to classification, emphasizing the need for time-sensitive, explainable AI approaches for personalized mental health screening.