基于随机森林的分类器在犯罪预测中的应用研究  

Application of an Improved Random Forest Based Classifier in Crime Prediction Domain

作　　者：孙菲菲[1] 曹卓[2] 肖晓雷 

机构地区：[1]中国人民公安大学反恐怖学院,北京100038 [2]中国人民公安大学犯罪学学院,北京100038 [3]中国人民公安大学侦查学院,北京100038

出　　处：《情报杂志》2014年第10期148-152,共5页Journal of Intelligence

摘　　要：犯罪预测一直是公安部门亟待解决的突出问题。基于随机森林这种模型组合分类器,结合机器学习技术在犯罪预测中的应用现状,提出了一种用于预测犯罪的新的分类方法,并通过模拟实验来展示这种分类方法比一般的随机森林分类会有更高的可信度。创新之处在于提出的这种随机森林分类器的每一棵树都是退化的决策树,并且根据在线学习的结果在下一轮的分类中选择区分度更高的决策树。最终给出一个应用于犯罪预测的较为成功的分类器的思路和模式,得出准确有效的预测结论。Crime prediction has always been an outstanding issue for public security department.Based on a a branch of Module Combination Classifier--Random Forest classifier,this article introduces a new way of classification which can be applied to crime prediction based on the current situation of the application of Machine Learning technology.We propose a novel Random Forest classifier which is composed of a group of degradation decision trees,and decision trees that are more distinctive are selected in the next round classification.Finally the article gives a successful new way of building classifier in crime predication domain,to reach an accurate and effective conclusion,and the high reliability of classifier results compared to general random forest classifiers is also demonstrated based on simulation experiment results.

关 键 词：随机森林 机器学习 分类器 犯罪预测 决策树 数据挖掘 

分 类 号：G353[文化科学—情报学]