Research On Key Person Crime Risk Prediction Technology Based On Fusion Data

Crime prediction with big data is an important measure for the Ministry of Public Security to implement the big data strategy and in-depth implementation of the smart police service,which plays a vital role in maintaining public safety.Existing crime researches mainly focus on inferring crime locations or criminal groups.However,the predicted crime locations or groups are too large to locate a specific person.Moreover,as individual offenders typically have no cooperative relationship with other offenders,it is difficult to apply existing methods on predicting crime groups to construct criminal network for individual crime.The known individual crime prediction study unilaterally uses personal basic information or trajectory,relies on manual design features and does not comprehensively use the two aspects of information,resulting in low prediction accuracy.So,this dissertation is dedicated to predicting individual crime risks rather than crime locations or groups,which comprehensively utilize GPS trajectories and personal basic information.After being authorized and anonymizing the data to ensure personal privacy,this dissertation collects a real-world dataset from a public security system and conduct studies.This dissertation proposes the crime risk prediction method based on personal fusion information graph.This method integrates trajectory and basic information by constructing the personal fusion information graph.Then,an individual crime risk prediction model is proposed,in which the self-attention graph pooling layer is used to filter the noisy trajectory nodes,the information fusion layer is designed to avoid the loss of personnel basic information during pooling process and to select the trajectory features.The experimental results show that on the evaluation metrics of1,it is 3.6%～12.8% higher than other methods,which proves the effectiveness of this method.Considering that the personal information fusion graph separates the information transmission among people and ignores the possible personal links,this dissertation further proposes a crime prediction method based on the joint fusion information graph which is constructed by using all person’s information.The joint fusion information graph builds a location-based social network by using the personlocation-person edge and the individual crime risk prediction is transformed into the classification of person nodes in the joint fusion information graph.Then,an individual crime risk prediction model based on joint information fusion graph is proposed.According to the characteristics of node connection and diffe rent information types in the graph,the graph attention networks under multiple relationships are designed,and the relationship level attention layer is used to aggregate person embeddings under different relationships.The experimental results show that on the evaluation index of 1,it is 8.4% higher than the crime risk prediction method based on personal fusion information graph,which further improves the accuracy of personal crime risk prediction.