Machine Learning Models For Information Support In The Justice System
Olha Ya. Kovalchuk, Vladyslav Teremeckyi, Andrii Kolesnikov, Natalia Chudyk, Valerii Kaniuka · 2024
The limited ability of people to efficiently and quickly process complex and diverse information from big data has led to the need for digital transformation, which has already become a reality for justice systems. The work presents an effective gradient-boosted decision tree model for predicting recidivism risk levels based on information about previous criminal activity, and demographic, and social characteristics of convicts. The dataset for applied research includes information about 13,010 individuals serving sentences in correctional facilities in Ukraine. Established that the number of previous convictions before serving sentences in correctional institutions and the number of suspended convictions imposed by court decisions in previous cases significantly influence the propensity of convicts to commit repeat offenses. The proposed machine learning model can be used to predict recidivism risk levels for new convicts, provide reliable information support for judicial decision-making, and become part of the judicial information system in Ukraine.