Ensemble Regression Tree with Bayesian Optimization for Prediction of Biochemical Oxygen Demand and Climate Impact Assessment in Full Scale Waste Water Treatment Plant

Wastewater treatment is crucial for pollution reduction and water quality promotion, but uncertainties in environmental conditions, influent shock, and wastewater treatment technology can impact effluent water quality, operation costs, and environmental risks. Machine Learning (ML) approaches can help reduce these challenges. Using a benchmark dataset, this paper examines the impact of hydraulic, climate-based, and wastewater characteristics parameters on the prediction of biochemical oxygen demand (BOD). BOD measures the amount of biodegradable organic matter that can be oxidized naturally by micro-organisms, indicating the amount of organic materials that must be eliminated. The proposed optimized prediction model (BO-ERT) is a predictive model that combines features from various sources using data engineering methods. It uses the Minimum Redundancy Maximum Relevance (mRMR) feature selection technique to identify the most relevant variables for training. The model aims to improve prediction accuracy while reducing computing costs. Bayesian optimization is used to optimize the hyperparameters of the ensemble regression tree technique, which can handle complex and non-linear interactions between variables. The proposed model outperforms linear regression, regression tree, and ensemble Bagged and Boosted trees in terms of accuracy on the used dataset. The model's feature selection phase reveals that specific variables, such as Chemical Oxygen Demand, Total Nitrogen, and Ammonia from wastewater features, as well as climate factors like humidity and wind speed, have the highest impact on predicting BOD levels. BO-ERT outperformed other regression methods and had a reasonable prediction error for the testing set, it achieved 0.58 coefficient of determination (R2), 43 Mean Absolute Error (MAE), and 58.4 Root Mean Square Error (RMSE). These results show that the model has potential usefulness in accurately predicting BOD levels in wastewater treatment plants.