Research On The Prediction Technology Of Criminal Information Based On Artificial Neural Network

Driven by the concept of intelligence-led police affairs,improving the analysis and judgment ability of criminal intelligence is the key link for public security organs to strengthen crime prevention and rational police decision-making.Therefore,it is particularly important to achieve accurate prediction of criminal intelligence.In recent years,with the arrival of the big data wave,data-driven crime intelligence prediction has become a hot topic in the academic community.Researchers have made some explorations in research directions such as crime situation prediction,prediction of the occurrence probability of specific crime types,prediction of crime probability of specific people,and prediction of crime types.However,the research in these directions is not enough,the prediction techniques proposed by the researchers are few,and the prediction accuracy of the prediction techniques needs to be further improved.In particular,the direction of crime type prediction is limited by experimental data--the only publicly useful San Francisco crime database available in this direction via our investigation.The research results in this direction are very scarce,and the accuracy of the prediction technology realized is relatively low,which cannot meet the actual combat demands of the public security.Therefore,we specifically collected 145,962 criminal cases involving 50 crime types from the public security department of H City,and formed a database with a total data volume of 1,605,582.This paper proposes two crime type prediction algorithms based on time-crime type count vectorization and dense neural network crime type prediction,and also based on the fusion of dense neural network and long & short-term memory neural network.In addition,we verified the advancement of these two algorithms on the San Francisco crime database and H city crime database we collected before.In terms of the problem that the existing crime type prediction algorithms do not fully exploit the relationship between time attributes and crime types,this paper proposes a crime type prediction algorithm based on time-crime type count vectorization and dense neural network.It mainly includes two steps: feature construction based on time-crime type count vectorization and prediction based on intensive neural network.Among them,the time-crime type vectorization technique can more fully explore the correlation between crime time and crime type.The designed and debugged intensive neural network can better fit the relationship between constructed features and crime types.For verifying the effectiveness and universality of the algorithm,we conducted experimental verification on H city crime database and San Francisco crime database.Aiming at the problem that the existing crime type prediction algorithm fails to predict the future crime type by using the existing data without considering the sequence of the cases,this paper proposes a crime type prediction algorithm based on the combination of intensive neural network and long & short-term memory neural network.Firstly,the intensive neural network and the short & long-term memory neural network were established and trained on the training set respectively.Then,input the output results of the two into a new neural network for training to achieve the fusion of the two neural networks.Thereinto,the use of intensive neural network can effectively fit the relationship between the constructed characteristics and crime types.The use of long & short-term memory neural network can excavate the logical relationship of criminal cases in time dimension.Similarly,the comparative experiment between the San Francisco crime database and the H city crime database proves the advancement of the algorithm.