Research on mental health states involves paying increasing attention to changes in daily life. Researchers have attempted to understand such daily changes by relying on self-reporting through frequent assessment using devices (smartphones); however, they are mostly focused on a single aspect of mental health. Assessing the mental health of a person from various perspectives may help in the primary prevention of mental illness and the comprehensive measurement of mental health. In this study, we used users' smartphone logs to build a model to estimate whether the scores on three types of questionnaires related to quality of life and well-being would increase compared to the previous week (fluctuation model) and whether they would be higher compared to the average for that user (interval model). Sixteen participants completed three questionnaires once per week, and their smartphone logs were recorded over the same period. Based on the results, estimation models were built, and the F-score ranged from 0.739 to 0.818. We also analyzed the features that the estimation model emphasized. Information related to "physical activity," such as acceleration and tilt of the smartphone, and "environment," such as atmospheric pressure and illumination, were given more weight in the estimation than information related to "cyber activity," such as usage of smartphone applications. In particular, in the Positive and Negative Affect Schedule (PANAS), 9 out of 10 top features in the fluctuation model and 7 out of 10 top features in the interval model were related to activities in the physical world, suggesting that short-term mood may be particularly heavily influenced by subjective activities in the human physical world.
Keywords: machine learning; mental health; quality of life; smartphone; well-being.
Copyright © 2022 Yao, Kaminishi, Yamamoto, Hamatani, Yamada, Kawada, Hiyama, Okimura, Terasawa, Maeda, Mimura and Ota.