Comparative Analysis of Machine Learning Models for Recidivism Prediction Based on Chi-square Test

Abstract

In order to excavate the influencing factors of recidivsim of the prisoners so as to achieve the purpose of prevention and redction of crime. This article proposes a feature selection method based on the experience of field experts and chi-square test, and uses the data from 2004 survey of inmates in state and federal correctional facilities as source, through data cleaning and data discretizes, and select five machine learning models for training and prediction respectively. Taking the accuracy rate, recall rate and values as evaluation indicators, compared the recidivism prediction capabilities of the five models. The results show that the feature selection method proposed in this paper can greatly impove the accuracy and recall rate of each model, and the logisitc regression model has a strong comprehensive ability.