A Machine Learning-Based Decision Support System Design for Restraining Orders in Turkey

Abstract

Restraining orders issued by the family court are among primary methods for fighting the ever-growing problem of domestic violence against women in Turkey. However, even before the cancellation of an inclusive law issued to protect women, the system failed to provide effective protection for the victims of violence. One of the main reasons for this failure is that there is no standardized risk assessment method used during and after the orders are issued, making it impossible to specialize them according to the changing needs of the victims. To solve this problem, the study aims to provide a framework for a decision support system that runs on previous criminal history and the offence records of the suspect. For indicating the system's feasibility, the public records on the inmates published by the Florida Department of Corrections are used where the results from the most effective model yielded a 64.4% ROC accuracy rate when classifying the type of the following crime that the offender will commit.