河北大学学报(自然科学版) ›› 2017, Vol. 37 ›› Issue (4): 426-433.DOI: 10.3969/j.issn.1000-1565.2017.04.015机器学习模型在预测服刑人员

• • 上一篇    下一篇

机器学习模型在预测服刑人员再犯罪危险性中的效用分析

马国富,王子贤,马胜利   

  • 收稿日期:2016-11-04 出版日期:2017-07-25 发布日期:2017-07-25
  • 作者简介:马国富(1974—),男,河北保定人,中央司法警官学院副教授,主要从事信息安全、机器学习方向研究. E-mail:magf2003@126.com
  • 基金资助:
    教育部人文社会科学研究规划基金项目(14YJAZH055);中央司法警官学院青年教师学术创新团队资助项目

Analysis of the effectiveness of machine learning model in predicting the risk of inmates

MA Guofu,WANG Zixian,MA Shengli   

  1. Department of Information Management, the National PoliceUniversity for Criminal Justice, Baoding 071000, China
  • Received:2016-11-04 Online:2017-07-25 Published:2017-07-25

摘要: 通过对国内外服刑人员的危险性及其再犯罪评估现状梳理,发现基于量表的传统服刑人员危险性评估工具的适应性和精确性越来越受到挑战.由数据和参数驱动的机器学习模型可以不断地进行自学习,从而不断地提高模型的适用性和精确性.首先对LR(logistic regression)、CART(classification and regression tree)、CHAID(chi-squared automatic interaction detection)、MLPNN(multi-layer perceptron neural network)4个常见的分类机器学习模型进行了介绍;在此基础上以2004年美国司法统计局(BJS)对服刑人员的调查(SISFCF)数据作为数据源,用灵敏率、特效率、准确率和AUC等评价指标对这4个模型进行了效用评估;最后对4个模型的预测能力进行比较.

关键词: 机器学习, 预测, 再犯罪, 危险性评估

Abstract: By analyzing the current situation of risk assessment of inmate at home and abroad, we find that the adaptability and accuracy of the traditional risk assessment tool of inmate based on the scale is being in creasingly challenged.However,the machine learning model driven by the data and parameter can be self learning,so as to continuously improve the applicability and accuracy of the model.Firstly, the paper introduces the four common machine learning models of LR, CART, CHAID and MLPNN; then,using the 2004 survey of inmates in state and federal correctional facilities(SISFCF)as the data source, the four models were evaluated by the sensitivity, specificity, accuracy, AUC and other evaluating indicators;finally, the predictive ability of the four models are compared.

Key words: machine learning, prediction, recidivism, risk assessment

中图分类号: