A Binary Logistic Regression Model for Support Decision Making in Criminal Justice

Abstract

Abstract Research background: The economics of incarceration is having an increasing impact on the economies of the world due to the rapid growth in the number of prisoners in the world The search for effective solutions that can help reduce government spending on prisoners in penitentiaries and at the same time ensure the safety of society is becoming increasingly important. These studies used the method of binary logistic regression to predict the probability of convicted criminal recidivism in the future. Purpose: The aim of the paper is to build an effective forecasting model that, based on the statistical and dynamic data of convicts, will provide information for optimal post-trial decisions, such as the grounds for possible parole, probation or length of sentence. Research methodology: The data were collected on the basis of statistical data of 13,010 convicts serving sentences in penitentiary institutions in Ukraine. To predict the probability of convicts committing criminal offenses binary logistic regression and ROC-analysis (Receiver Operator Characteristic analysis) were used. Results: A qualitative binary logistic regression model has been constructed, with the help of which it is possible to predict the probability of criminal recidivism by each of the convicts on the basis of its individual values of the variables included in the model. Novelty: For the first time in Ukraine, a model has been developed to predict the probability of convicts committing repeated criminal offenses.