[4F3-GS-3] Knowledge utilization and sharing:

[4F3-GS-3-02] Learning Decision Tree with Latent Variables for Mortality Analysis on Mental Illness

〇Yuta Shikuri^1,2, Kei Tokutsu¹, Kenji Fujimoto¹, Shinya Matsuda¹ (1. University of Occupational and Environmental Health Japan, 2. Tokio Marine Holdings, Inc.)

Keywords:Mortality Analysis, Bayesian Network, Decision Tree

The analysis of relations between mental illness and mortality could potentially lead to a reduction in the mortality rate among patients with mental illness. However, assessing the status of mental illness by clinical examination results is challenging. Therefore, we consider performing mortality analysis based on the combination of prescription drugs while taking into account the computational complexity.In this study, we propose a method to capture the potential status of mental illness through drug prescription patterns using a Bayesian network. Experimental results with medical receipt data demonstrate that candidate branching conditions based on this potential status improved the decision tree learning score for a comparable execution time relative to an existing method. Our approach can be used for the analysis in situations where the combinations of multiple observed variables represent potential states that influence outcomes.