基于多模态信息特征融合的犯罪预测算法研究

RESEARCH ON CRIME PREDICTION ALGORITHM BASED ON MULTIMODAL INFORMATION FEATURE FUSION

近年来,对犯罪预测方法进行了很多研究,这些方法在犯罪预测中存在高度非线性关系、冗余和多个数据集之间的依赖关系等问题。为了提高犯罪预测模型正确率,设计一种多模态信息特征融合的犯罪预测算法。将空间、时间、环境和上下文信息特征融合提高群集聚类正确率,充分利用特征目标函数计算逻辑损失,从而提高全局特征的正确率和局部特征的精确率。实验结果表明,在不同比例数据训练集条件下,该算法比现有方法的正确率和精确率分别提高了约12%和4%。

In recent years,many researches have been conducted on crime prediction methods. These methods have problems such as highly nonlinear relationships,redundancy,and dependencies among multiple data sets in crime prediction. In order to improve the accuracy of crime prediction model,a multi-modal information feature fusion crime prediction algorithm was designed. The algorithm fused spatial,temporal,and environmental and context information features to improve clustering accuracy. It made full use of the characteristic objective function to calculate the logic loss,thereby improving the accuracy of global features and the accuracy of local features. The experimental results showed that the accuracy and precision of the proposed algorithm were improved by about 12% and 4% respectively compared to the existing methods under the conditions of different ratio data training sets.