Enhancing Explainability of Chemical Oxygen Demand Prediction Models for Industrial Wastewater Treatment in Delfzijl, the Netherlands

Wastewater treatment is a vital solution that will reduce environmental effects by removing unwanted organic waste and producing a sustainable water source. However, the fundamental issue in running a wastewater treatment facilities is the excess chemical oxygen demand concentration due to sludge thickening. The quality of the WWTP’s effluent will decrease due to sludge thickening as a result of excess concentration of COD which will lead to decrease treatment efficiency, increased operational challenges, environmental harm to high aquatic life, and unpleasant odours. Consequently, one of the crucial problems that must be resolved to guarantee process safety and enhance effluent quality is the accurate identification of COD concentration. Machine learning, a branch of artificial intelligence (AI), is regarded as having the potential to disrupt global economies, environments, and communities by enabling systems to learn directly from data, examples, and experience without the need for pre-established rules. This research work made extensive analysis the estimated the concentration of COD level of Industrial Wastewater Treatment Plant which is located in Delfzijl City of the Netherlands, using sophisticated machine learning such as CatBoost model, XGBoost model and Random-Forest model for cost effectiveness and efficiency improvement of wastewater treatment facilities. The models demonstrated their capacity of handling the complexity in the dataset with prediction accuracy of 0.98.Additionally, to enable interpretation and highlight the features that contributed the most in the output of the models’ prediction, the SHAP explainer called SHapley Additive exPlanations was added and demonstrated a very good performance.