基于贝叶斯优化轻量级梯度提升机模型预测淡水鱼中恩诺沙星抽检结果

Prediction of enrofloxacin sampling inspection results in freshwater fish based on Bayesian optimization light gradient boosting machine model

中文关键词:  淡水鱼  恩诺沙星  贝叶斯优化  轻量级梯度提升机  重采样  预测模型

英文关键词:freshwater fish  enrofloxacin  Bayesian optimization  light gradient boosting machine  resampling  predictive model

中文摘要:

      目的 建立基于贝叶斯优化轻量级梯度提升机(Bayesian optimization light gradient boosting machine, BO-LGBM)的预测模型, 对上海市市售淡水鱼中恩诺沙星的抽检结果进行预测。方法 基于2020—2023年上海市市售淡水鱼中恩诺沙星的抽检数据, 以抽样地区、抽样月份、抽样环节、抽样场所、检测机构、淡水鱼品种6个特征为输入变量, 以淡水鱼中恩诺沙星的抽检结果是否合格为输出变量, 采用十折交叉验证法从7种机器学习算法中筛选出的最佳模型作为初始模型, 逐步构建BO-LGBM模型及其与5种数据重采样方法结合的预测模型。结果 上海市市售11114件淡水鱼中恩诺沙星的不合格率为4.7%。其中, 泥鳅中恩诺沙星的不合格率最高(27.1%), 其次为鳝鱼(24.6%)和鳊鱼(18.8%)。相比较其他模型, 采用自适应合成抽样(adaptive synthetic sampling, ADASYN)与BO-LGBM结合模型对测试集样品抽检结果预测的准确率(0.83)、召回率(0.86)、F1值(0.82)和AUC值(0.84)均为最高。结论 ADASYN-BO-LGBM模型的预测性能好, 能够较为准确地预测淡水鱼中恩诺沙星抽检结果, 为基于问题导向的淡水鱼监督抽检方案制定及风险预警提供技术支撑。

英文摘要:

      Objective To establish a prediction model based on Bayesian optimization light gradient boosting machine (BO-LGBM) for predicting the sampling inspection results of enrofloxacin in freshwater fish sold in Shanghai. Methods Based on the sampling inspection data of enrofloxacin in freshwater fish sold in Shanghai from 2020 to 2023, the 6 kinds of characteristics included sampling area, sampling month, sampling process, sampling site, testing institution, and freshwater fish species were taken as input variables, and whether the sampling inspection results of enrofloxacin in freshwater fish were qualified as the output variables. The best model selected from 7 kinds of machine learning algorithms used ten fold cross validation was used as the initial model, and the BO-LGBM model and the prediction model combined with 5 kinds of data resampling methods were gradually constructed. Results The unqualified rate of enrofloxacin in 11114 pieces of freshwater fish sold in Shanghai was 4.7%. Among them, the unqualified rate of enrofloxacin in Misgurnus anguillicaudatus was the highest (27.1%), followed by Monopterus albus (24.6%) and Parabramis pekinensis (18.8%). Compared with other models, the adaptive synthetic sampling (ADASYN) combined with BO-LGBM model had the highest accuracy (0.83), recall (0.86), F1 value (0.82), and AUC value (0.84) in predicting the sampling inspection results of the test set samples. Conclusion The ADASYN-BO-LGBM model has the better predictive performance and can accurately predict the sampling results of enrofloxacin in freshwater fish, which can provide technical support for problem-oriented freshwater fish supervision sampling plan formulation and risk warning.