Binary-Relevance Classification of Depression and Anxiety in the Elderly Using Low-Cost Activity Trackers

Abstract

Understanding mood disorders of depression and anxiety in the elderly is important because those are common symptoms of dementia. This paper shows our methodology of supervised learning to understand the relationship between the diagnosis of depression and/or anxiety of the elderly and their activity patterns. Using low-cost activity trackers, activity data of the elderly were collected over several weeks per subject. We extracted features from the time-series activity data based on circadian rhythms. We have applied binary relevance to this problem of multi-label nature diagnosed with depression, anxiety or both. We tried various classification algorithms for each label and compared the performances to find the best classifier among the algorithms. The result yielded that the best classifier was the F-1 score of 77.4% in the classification. Our study has yielded a meaningful result in screening mood disorders of the elderly using low-cost activity trackers.