Depression is one of the mood disorders that over 300 million suffer from.
This master thesis carried out prediction as well as one-year studies to find lifestyle habits and physical constitution related to depressive tendencies. In the prediction study, depressive tendencies were predicted from lifestyle habits and physical constitution using machine learning. The predictive performances were evaluated, followed by finding important features under two conditions. That is, we used (1) balanced (same number of participants with high or low depressive tendencies) and imbalanced data which is close to the real situation (few participants with high depressive tendencies), and used (2) all questions and questions, deemed unrelated to depressive states. In the imbalanced data, an under sampling and bagging approach was applied to improve predictive performances. The result was that highest predictive performances were obtained when using all questions in the balanced data. Important features were found from variety of categories, physical constitution, dietary habits, and health condition. In the one-year study, we attempted to find lifestyle habits and physical constitution that co-vary with the change in depressive tendencies through certain periods by Jensen-Shannon divergence. The finding from the study was that some lifestyle habits such as consumption of deep yellow vegetables and physical constitution, feeling arms and legs heavy co-varied with the change in depressive tendencies.