Predicting Recidivism Through Machine Learning

• In Event: Advances in Machine Learning and Crime

Abstract

This study utilizes contemporary classification techniques, such as machine learning and explores associations (known as knowledge discovery) in a relatively large dataset of recidivism. Specifically, random forests, recursive portioning, support vector machines, and classifiers, such as naïve Bayes and K-nearest neighbors will be used for statistical learning, whereas the least absolute shrinkage and selection operator (LASSO) and VIF regression methods will be used for knowledge discovery. This study will attempt to provide answers to questions such as do actuarial models outperform traditional approaches?

Author

• Turgut Ozkan, University of Texas at Dallas