Prediction of population behavior of Listeria monocytogenes based on ComBase database

ObjectiveA hybrid model based on bidirectional short term memory （BiLSTM） and Transformer was constructed to predict the population behavior of Listeria monocytogenes under different environmental conditions.MethodsIn view of the limitations of traditional machine learning methods in processing complex time series data， an innovative solution combining BiLSTM and Transformer model is proposed to effectively capture the long and short term dependence of time series and improve the prediction accuracy. The model input includes characteristics such as temperature， water activity， pH value， time and whether it is the initial bacterial concentration， etc. After data preprocessing， feature standardization and category coding， the trained model is used for prediction. The experimental data of L. monocytogenes from ComBase database were used to verify the model.ResultsThe model performed well for several food groups， including beef， pork， medium， seafood and fruits， with R² values of 0.72， 0.65， 0.85， 0.81 and 0.81， and with RMSE values of 1.17， 1.15， 0.89， 0.93 and 0.83， respectively. The results showed that the model could accurately capture the changing trend of bacterial population. By calculating the coefficient of deviation （Bf） and accuracy （Af）， the advantages of the model in forecasting accuracy and robustness were both verified.ConclusionThe BiLSTM-Transformer hybrid model as an efficient and accurate method for predicting bacterial population behavior， can contribute to bacterial prediction in the field of food safety.