[2P5-J-2] Machine learning: medicine and heathcare

[2P5-J-2-01] Prediction of depressive tendencies from health data considered unassociated with depressive states

Shiori Yamaguchi¹, 〇Hiroki Tanaka¹, Hayato Maki¹, Shigehiko Kanaya^1,2, Nobutaka Suzuki³, Satoshi Nakamura^1,2 (1. Nara Institute of Science and Technology, Graduate School of Science and Technology, 2. Nara Institute of Science and Technology, Data Science Center, 3. Kanazawa University, Graduate School of Medical Sciences)

Keywords:Machine Learning, Depressive tendencies, Health data

This study investigated firstly, which lifestyle habits and physical constitution are considered related to depressive states. Following the exclusion of the lifestyle habits and physical constitution that are related to depressive states, we examined that whether the rest can predict depressive tendencies.
We used three classifiers for the prediction. In addition, we used a combination of undersampling and bagging approach to improve predictive performances in our imbalanced data.
The finding was that we obtained recall of 0.65 from the logistic regression with L2 regularization. Moreover, some important questions for the prediction were included in DSM-5 or QIDS-SR, meaning that we found that not all main symptoms have been widely known.