This study aims to compare and verify the predictive performance of logistic regression analysis and decision trees, which have been widely used in recent years for estimating perpetrator profiles. While differences in predictive performance when estimating a perpetrator's criminal history have been examined, other attributes have not been sufficiently investigated.
This study focuses on single-hit murder cases in Japan, using case information collected from a newspaper database to conduct logistic regression and decision tree analyses with the perpetrator's employment status as the dependent variable. Estimation accuracy is measured using one-miss cross-validation (LOOCV), and AUC and discrimination accuracy are used as evaluation metrics for comparison. In addition, the explanatory variables selected for each method are compared to explore factors that contribute to the estimation of employment status.
