Detection of Temporal Changes in the Behavior of Older Adults Living Alone by Means of Continuous Data Collection in the Home

Background: In Japan, the number of older adults living alone has been increasing, raising the risk of unnoticed health decline or solitary death. Continuous monitoring using sensors can help detect behavioral changes indicating health issues and has the potential to support both older adults and their families. Methods: We obtained behavior and temperature data, continuously recorded over a long period at 15-min intervals from sensors installed in the homes of nine older adults living alone. After data cleaning, behavioral signals were analyzed using Fourier spectral analysis and multiple regression to extract 13-dimensional behavioral characteristic vectors. We whitened a portion of these behavioral characteristic vectors as benchmark data. We applied the same whitening process to the comparing data using the matrix obtained during this whitening process. By analyzing misclassification rates using boundary variance for benchmark and comparing data, we attempted to detect temporal changes in user behavior and differences between individuals. Results: Spectral analysis revealed 24-hour periodicity in all users’ behavior. By analyzing the misclassification rate using boundary variance for long-term signals, we identified users who maintain consistent behavioral patterns and those exhibiting significant temporal variation. We were also able to detect differences in these behavioral patterns. Conclusions: This study demonstrates that long-term temporal changes in the daily behavior of older adults living alone can be detected using simple continuous sensor data. Our approach is applicable not only for monitoring behavior changes in older adults living alone, but also for observing behavior changes in people with disabilities and children within the home environment.