Construction and Evaluation of Intimate Partner Homicide Prediction Model.

Objectives: To analyze the independent influencing factors of intimate partner homicide (IPH) cases, construct an IPH prediction model, and provide a basis for criminal profiling.

Methods: A total of 476 convicted homicide cases in Guangdong Province from January 1, 2014, to December 31, 2020, were collected as modeling dataset. They were divided into the IPH group (n=180) and the non-intimate partner homicide (N-IPH) group (n=296) based on whether the offender and victim were intimate partners. Logistic regression was used to build the model, the model was evaluated through the receiver operating characteristic (ROC) curve analysis and a nomogram was drawn. Internal validation was conducted using ten-fold cross-validation method. A total of 126 court judgments from outside Guangdong Province from January 1, 2011, to December 31, 2020, were randomly collected for external validation.

Results: Through multi-factor Logistic regression analysis, 7 variables were ultimately selected for inclusion in the model. The Hosmer-Lemeshow goodness of fit test result of the model was χ²=13.158, P=0.068. The ROC area under the curve (AUC) of the model was 0.939 (95% CI: 0.919-0.959), the cut-off value was 0.292, the sensitivity was 0.900, and the specificity was 0.865. The calibration curve was close to the ideal curve. The ten-fold cross-validation showed the accuracy of 0.863 and a Kappa value of 0.708. The external validation results showed an AUC of 0.922 (95% CI: 0.872-0.971), a cut-off value of 0.292, a sensitivity of 0.890, and a specificity of 0.886. The calibration curve tended to the ideal curve.

Conclusions: The IPH prediction model based on forensic field indicators has good predictive ability, reliable accuracy and stability, and can provide a scientific method for criminal profiling.