基于数据挖掘的兽药质量风险预测

Quality-Risk Prediction of Veterinary Drugs by Data Mining

摘要:

当前，兽药残留已成为食品安全的源头问题之一，养殖户缺乏鉴别假兽药的能力，兽药质量风险较大，为了提高养殖户的辨假能力，减少不合格兽药的使用，降低兽药使用风险，通过整理中国兽医药品监察所的抽检数据，运用SPSS Modeler软件，以C5.0、Logistic、神经网络构建数据挖掘的分类预测模型，对兽药质量进行分类预测。发现三种分类模型的整体分类精度偏低，对此，选用组合分类器对模型进行了优化，并对神经网络、二元逻辑回归-神经网络及决策树-神经网络进行了比较，发现从分类精度以及泛化性能上来讲，决策树-神经网络的整体表现最好，最后，本文构建了决策树-神经网络的兽药质量风险预测模型，并对之进行了进一步的优化，预测准确率能达到74.34%，可为养殖户的购买决策提供参考。

Abstract:

Veterinary drug residues had become one of the source problems for food security at present. It was difficult for farmers to identify the fake veterinary drugs, which resulted in the risks of veterinary drugs quality. To improve the identification ability of the farmers and reduce the utilization of unqualified veterinary drugs, the data mining classification prediction model, established by C5.0, Logistic, neural network, was used to classify and predict the quality of veterinary drugs by sorting the sampling data of Chinese Veterinary Drug Administration based on SPSS Modeler software. Results showed that the classification accuracy of the three models was low, which resulted in optimizing the model by combination of classifier, and the neural network, binary logic regression - neural network, decision tree-neural network were compared. The overall performance of decision tree - neural network was the best in classification accuracy and generalization performance. Finally, the model for predicting the veterinary drugs quality in decision tree-neural network was established and further optimized, and the prediction accuracy reached 74.34%..