A Feature Selection Method Based on Graph Coding for Detention Risk Assessment in Criminal Cases

Abstract

The invention discloses a feature selection method for criminal custody risk assessment based on graph coding combined with deep reinforcement learning, which is mainly used for important feature selection of criminal detention risk assessment. The present invention constructs a criminal feature knowledge map by introducing an external knowledge base, and then uses the graph attention network layer and multi-graph interaction to encode the graph, and uses multi-task prediction combined with deep reinforcement learning to infer the necessity of detention, and finally encodes the graph according to the feature map Part of the attention distribution selects features with higher weights to complete the task of feature selection for custody risk assessment. Since the criminal characteristics of suspects recorded in different criminal cases are different, this method conducts targeted training on different criminal cases to improve the accuracy of the model. The criminal cases handled by this method include: crime of theft, crime of dangerous driving, crime of causing traffic accident, crime of fraud, crime of intentional injury, crime of robbery, crime of rape, crime of accommodating others to take drugs.

Description

technical field [0001] The invention belongs to the field of natural language processing, and relates to a feature selection method for criminal case custody risk assessment based on graph coding. Background technique [0002] With the rapid development of machine learning, various machine learning algorithms are widely used in more and more fields, and the results of these machine learning have brought great convenience to people. In machine learning, feature engineering is the top priority, and selecting appropriate features will improve the performance of the model. More specifically, selecting better features can bring the following benefits to machine learning models: [0003] 1) Reduce the complexity of the model, saving a lot of computing resources and computing time; [0004] 2) Improve the generalization ability of the model. Generalization ability refers to the adaptability of machine learning algorithms to fresh samples. In layman's terms, if a model has a goo...

AI Technical Summary

Problems solved by technology

However, as the statistics on the characteristics of criminal suspects become more and more comprehensive, and the consideration of crime types and situations becomes more and more specific, the difficulty of feature selection gradually increases.

Images