The rate of recidivism by criminals after their release from prison is high, which is harmful to society. Thus, it is socially significant to reduce their recidivism rate. This article uses public data from the state of Iowa in the United States. According to the data characteristics, such as having redundant samples and mixed attributes, we propose the following methods. First, we use a rough set attribute reduction algorithm based on probability distributions to reduce the redundant items. Second, the sample data are clustered with an improved clustering algorithm. Based on the traditional K-prototype clustering algorithm, the clustering algorithm is improved by changing the measurement method of the categorical attributes, changing the initial cluster center selection method, and weighting the attributes based on the information entropy. The clustering experiment results show that the improved clustering algorithm has a better clustering effect and higher clustering accuracy than the traditional K-prototype clustering algorithm. Finally, a back propagation neural network is used to predict the recidivism probability of the sample processed by the above algorithm. The final experimental results show that the two redundant attributes are successfully reduced by rough sets, which greatly reduces the run time of the model. Compared with the traditional K-prototype clustering algorithm, the improved K-prototype clustering algorithm proposed in this paper has a better effect on the various indicators and objective function. Finally, through neural network prediction, the prediction accuracy of this model reached 87.9%. At the same time, a large number of experiments on benchmark datasets verify the effectiveness of our proposed model.

罪犯出狱后再犯率高，危害社会。因此，降低他们的再犯率具有社会意义。本文使用来自美国爱荷华州的公开数据。根据数据特征，如具有冗余样本和混合属性，我们提出以下方法。首先，我们使用基于概率分布的粗糙集属性约简算法来减少冗余项。其次，使用改进的聚类算法对样本数据进行聚类。该聚类算法在传统K-prototype聚类算法的基础上，通过改变分类属性的度量方法、改变初始聚类中心选择方法、基于信息熵对属性进行加权等方式进行改进。聚类实验结果表明，改进后的聚类算法比传统的K-prototype聚类算法具有更好的聚类效果和更高的聚类精度。最后，利用反向传播神经网络对上述算法处理后的样本再犯概率进行预测。最终的实验结果表明，粗糙集成功地减少了两个冗余属性，大大减少了模型的运行时间。与传统的K-prototype聚类算法相比，本文提出的改进K-prototype聚类算法在各项指标和目标函数上都有更好的效果。最后，通过神经网络预测，该模型的预测准确率达到了87.9%。同时，