食品工业科技2025，Vol.46Issue(6)：43-55,13.DOI:10.13386/j.issn1002-0306.2024050005
ED-Stacking:一种基于集成学习的小样本牛肉微生物生长预测模型构建方法
ED-Stacking:A Construction Method of Few-shot Prediction Model for Beef Microbial Growth Based on Ensemble Learning
摘要
Abstract
Under the current technological conditions,microbial detection was complicated and time-consuming,which leaded to the problem of lagging detecting results and limited sample size.In this paper proposed a construction method of few-shot predictive model for microbial growth in beef,called ED-Stacking,which was based on time series decomposition and ensemble learning,for early warning of microbial risks in food.First,empirical mode decomposition(EMD),discrete Fourier transform(DFT)and additive modeling were applied to construct a time series decomposition method EMD-DFT,which was used to extract the trend,period,and residual features in the microbial growth time series,and to provide training data for the subsequent prediction model.Second,these feature data were then utilized to train a single-layer linear neural network(SLN),extreme gradient boosting(XGBoost)and gradient boosting regression tree(GBRT).Finally,the stacking method in ensemble learning was used to fuse the three trained models to form ED-Stacking,a microbial growth prediction model with better performance in prediction.Results showed that ED-Stacking achieved 0.229 and 0.147 in MAE and MSE metrics,respectively,with lower prediction errors than the five baseline models of SLN,XGBoost,GBRT,GRU,and Transformer.Based on this model,the food quality classification was performed and the weighted precision of the classification,Weighted-Precision,reached 98.80％.Furthermore,the study also presented FMPvis,a visual analysis system for the prediction of microbial growth in food,which could display the prediction results and the food quality classification results,and helped users to analyze the trend of each environmental factor over time and its influence on the prediction results,so as to facilitate risk analysis and early warning.This approach contributes a new idea for early warning of microbial risk in food.关键词
食品安全/微生物生长预测/时间序列分解/集成学习/可视分析Key words
food safety/microbial growth prediction/time series decomposition/ensemble learning/visual analysis分类
轻工纺织引用本文复制引用
李汉强,陈谊,高宇飞,侯堃,宋丽萍,李静..ED-Stacking:一种基于集成学习的小样本牛肉微生物生长预测模型构建方法[J].食品工业科技,2025,46(6):43-55,13.基金项目
国家重点研发计划(2022YFF1100905) （2022YFF1100905）
国家自然科学基金项目(U23B2009). （U23B2009）
