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

发布时间：2018-07-31 17:27

【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.
[Abstract]:Through combing the status quo of the risk and reoffending assessment of the prisoners at home and abroad, it is found that the adaptability and accuracy of the traditional risk assessment tools based on the scale are being challenged more and more. The machine learning model driven by data and parameters can continuously self-learn and improve the applicability and accuracy of the model. This paper first introduces LR (logistic regression) cart (classification and regression tree) Chaid (chi-squared automatic interaction detection) / MLPNN) (multi-layer perceptron neural network) four common classification machine learning models, and then uses the (SISFCF) data collected by (BJS) of the US Judicial Bureau of Statistics in 2004 as the data source, and uses the sensitivity and efficiency as the data source. The utility of the four models is evaluated by accuracy and AUC, and the prediction ability of the four models is compared.
【作者单位】：中央司法警官学院信息管理系;
【基金】：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
【分类号】：D917;TP181

【相似文献】