Revised ANN Model to Predict Escherichia coli Classes in Lake Michigan Beaches

Lake Michigan beaches are monitored during beach-goer season every day, by the Indiana Department of Environmental Management (IDEM). Based on Escherichia coli results, beach closures are made. During 2019, artificial neural network based E. coli prediction tool was created to identify the E. coli classes using 13 input parameters observed together with E. coli. For this now-cast model, two classes were considered for the final model output [(1) for safe level with E. coli counts less than 235 CFU/100 mL, and (2) for E. coli counts more than 235 colony forming units/100 mL]. This year, IDEM recommended to update this model with additional data observed during 2020 through 2022. All the 13 inputs (including water temperature, pH, total dissolved solids, total suspended solids, turbidity, trashes, bird counts, color, odor, electrical conductivity) used in the 2019 model were collected from the beaches together with E. coli data during 2020–2022. This research paper presents the phase II results. After initial data consolidation, updated database had 761 data for the considered five beaches. After examining several architecture types for the feedforward neural network model, the best model was identified. The model was cross-validated by dividing the data into four blocks. New model results provided an overall predicting success of 93% for the two classes considered. However, the class 1 and class 2 prediction accuracies decreased from the original model developed using 2019 data. Class 1 and class 2 were predicted correctly 93% and 79%, respectively. It resulted in a 10% decrease in class 2 predictions. In this research work, a systematic review was conducted of the various ANN model results in different development stages, and the performance was examined for each year.