Biological oxygen demand prediction using artificial neural network and random forest models enhanced by the neural architecture search algorithm

The most critical wastewater quality indicators (WQIs) for diagnosing the performance of wastewater treatment plants are the biological oxygen demand (BOD) and chemical oxygen demand (COD). Measuring the five-day of biological oxygen demand (BOD5) level in wastewater requires significant consumption of energy and. This research unprecedentedly develops a new hybrid machine learning (ML) technique to predict BOD₅ based on the neural architecture search (NAS) algorithm coupled with deep neural network (DNN) and random forest regression (RFR) models for the first time. In order to calibrate and validate the proposed models, various wastewater quality variables including wastewater potential hydrogen (pH), specific conductance (SC), total suspended solids (TSS), and COD were selected. The performance accuracy of these hybrid models was compared with traditional multilayer perceptron neural network (MLPNN), RFR and multiple linear regression (MLR) models. The results have been compared and evaluated based on graphical inspection (Boxplot, Violin plot, Spider plot, and Taylor diagram) and statistical indicators such as correlation coefficient (R), Willmott's index of agreement (WI), root mean square error (RMSE) and mean absolute error (MAE). Notably, our study demonstrated that the accuracy of BOD5 prediction increased using only pH, SC, TSS and COD. It can also be concluded that the best accuracy was obtained using the NAS-RFR with an R, WI, RMSE and MAE of 0.953, 0.967, 4.775mg/L and 2.944mg/L, respectively at the Baraki plant and using the NAS-DNN model with an R, WI, RMSE and MAE of 0.934, 0.953, 1.886 mg/L and 1.400 mg/L, respectively at the Reghaia plant. Our results underline the promising potential of the NAS-DNN and NAS-RFR hybrid models for accurate prediction of BOD5 in Algeria. This model, which outperforms traditional models, can significantly help decision-makers in wastewater treatment plants and water quality indices.