Method Of Probability And Consequence Prediction Of Property Crime Incidents Based On Machine Learning

At present China witnesses high crime rates and low detection rates of property crime,which extremely threatens the property security of citizens.Thus,objective and timely risk analysis for property crime is quite necessary.As is known,risk is usually expressed in terms of a combination of the consequence and likelihood.However,traditional qualitative risk analysis methods of property crime are costly and difficult to be validated;traditional quantitative and semi-quantitative risk analysis methods of property crime are often considered Inaccurate and inflexible.In this study,method based on machine learning is proposed to predict the consequence and probability of property crimes including robbery,non-violent robbery and larceny.In order to predict the consequence of property crime incidents,characteristics“time of case incidence”,“region of case incidence”,“time of choice”,“place of choice”,“object of choice”,“per capita GDP”and“average monthly salary of employees”are extracted from a statistical yearbook dataset of ZS city from 2008 to 2014,and model of consequence prediction of property crime incidents based on machine learning is established.The results show that the gradient boosting decision tree(GBDT)algorithm has the best performance,and the prediction accuracy in terms of larceny consequence reaches 0.88.As for robbery and non-violent robbery,common and serious cases are more likely to occur in the prosperous areas,and severe cases are likely to occur in other places(excluding residential areas,remote areas,highway areas and prosperous area).Property crime incidents are more likely to occur in urban areas on the weekdays,and consequence levels are mostly common(the consequence is divided into three levels:common,serious and severe).In order to predict the probability of property crime incidents,features of crime attributes and weather conditions based on crime data and weather data of ZS city from February 1,2005to July 31,2015,are extracted.Results of property crime incidents prediction based on time lag model show that K nearest neighbor model has the best performance.R~2 value in terms of larceny reaches 0.83,and those of robbery and non-violent robbery reaches 0.88and 0.80,respectively.Results of property crime incidents prediction based on real-time data show that the K nearest neighbor regression model in terms of robbery has the best performance,with a R~2value of 0.7.The model can predict and analyze the consequence and probability of property crimes.The model can be applied to prevent and crackdown property crime,and provide decision-making support.