SOCIO-ECONOMIC DETERMINANTS OF RECIDIVISM. SOME PROBLEMS OF IDENTIFICATION RELATIONSHIPS USING QUANTITATIVE METHODS

Abstract

The aim of the author was to discuss an application of data mining and statistical methods to recidivism prediction. There was analysed a binary classification problem where the goal was to predict if a prisoner will be arrested for a certain type of crime within one year of being released from prison. There were compared different models such as neural network, classification tree, logistic regression and SVM. General accuracy of all the models exceeded 70% correctly classified instances, but all of the analysed classifiers were characterized by high “false negatives” ratio and so they would be useless in practice.

Keywords

• risk of recidivism
• survival analysis
• logistic regression
• data mining