Research On Individual Pattern Mining And Identity Prediction Of Property

Thanks to the reform and opening up policy,China has achieved tremendous economic achievements and created huge material wealth.Currently,the public security situation of China is also facing a new challenge.The challenge is that the property crimes are increasing with the development of economy and account for a large proportion that are to illegally occupy other people's property,which is a threat for urban life and safety.It is helpful for police departments to dig out the pattern of property crimes to prevent and combat crimes accurately in advance.The current research mainly focuses on the temporal-spatial pattern of criminal activities and the characteristics of criminals,which is certain deficiencies in the depth.One is lack of research on the pattern of criminal activity on special dates,the temporal trend prediction of crime-making activities of offenders and the temporal-spatial pattern of crime committing.The other is that many researches have adopted non-real-time data for police,such as psychology,income,emotion,pressure and so on,which is not conducive for police to predict the criminal identity in the application of police activities.Taking the theft of electric bicycles in Beijing as an example,we have carried out the following work in the analysis and prediction of the time pattern of criminal behaviors and the prediction of the identity of offenders.Firstly,we analyze the temporal patterns of criminals committing crimes at periodic and special date.The STLFNN model is proposed based on a seasonal trend decomposition procedure based on loess and Full Connected Neural Network,which is designed to predict daily theft level in order to find out when criminals are most likely to commit crimes.The prediction results show that this model perform significantly better than the traditional algorithm,such as ARIMA,Holt-Winters,LSTM,Prophet and so on.This model can realize the long-term prediction of the criminal's criminal behavior for a period of one year.Secondly,we have analyzed the spatial-temporal pattern of criminal's behaviors,based on the time and space factors.The machine learning methods are used to predict the identity information of offenders such as region,education,occupations and so on,which is on the basis of spatial-temporal pattern of criminal's behaviors and combined with the technique and the object of criminal activities.The prediction results show that the accuracy of the model can reach 80% for predicting the criminal's geographical identity.This paper could provide a method and practical example for the intelligence analysis and crime pattern analysis for policemen.We hope that this work could help police to investigate more effectively and crack down on crimes more precisely.