Comparative Analysis of Machine Learning Models for Time-Series Forecasting of Escherichia Coli Contamination in Portuguese Shellfish Production Areas

Shellfish farming and harvesting have experienced a surge in popularity in recent years. However, the presence of fecal bacteria can contaminate shellfish, posing a risk to human health. This can result in the reclassification of shellfish production areas or even prohibit harvesting, leading to significant economic losses. Therefore, it is crucial to establish effective strategies for predicting contamination of shellfish by the bacteria Escherichia coli (E. coli). In this study, various univariate and multivariate time series forecasting models were investigated to address this problem. These models include autoregressive integrated moving average (ARIMA), vector autoregressive (VAR), and long short-term memory (LSTM) networks. The data used for this study consisted of measurements of both E. coli concentrations and meteorological variables, which were obtained from the Portuguese Institute of Sea and Atmosphere (IPMA) for four shellfish production areas. Overall, the ARIMA models performed the best with the lowest root mean squared error (RMSE) compared to the other models tested. The ARIMA models were able to accurately predict the concentrations of E. coli one week in advance. Additionally, the models were able to detect the peaks of E. coli for all areas, except for one, with recall values ranging from 0.75 to 1. This work represents the initial steps in the search for candidate forecasting models to help the shellfish production sector in anticipating harvesting prohibitions and hence supporting management and regulation decisions.