Forecasting bacteriological presence in treated drinking water using machine learning

A novel data-driven model for the prediction of bacteriological presence, in form total cell counts, treated water exiting drinking treatment plants is presented. The was developed and validated using a year hourly online flow cytometer data from an operational plant. Various machine learning methods are compared (random forest, support vector machines, k-Nearest Neighbors, Feed-forward Artificial Neural Network, Long Short Term Memory RusBoost) different variables selection approaches used to improve model's accuracy. Results indicate that could accurately predict counts 12 h ahead both regression classification-based forecasts—NSE = 0.96 best model, K-Nearest Neighbors algorithm, Accuracy 89.33% classification combined random K-neighbors RusBoost algorithms. This forecasting horizon sufficient enable proactive interventions processes, thereby helping ensure safe water.