Enhanced Wastewater Parameter Prediction using Comprehensive Regression Framework: Leveraging Feature Engineering and Hyperparameter Optimization with Machine Learning

The amount of wastewater generated is depends on various factors such as population growth, industrial activities, and infrastructure development. Wastewater discharged without treatment can adversely impact the environment. This paper aims to design the regression models, train and validate their performance, and interpret the results in the context of wastewater treatment processes. This paper presents a ML based framework to predict BOD, COD, and TSS using various regression models. The framework consists of anomaly handle algorithms (Isolation Forest, One-class SVM and K-means clustering), feature selection techniques (SelectKBest, SelectFromModel, and Recursive feature elimination), hyperparamter techniques (grid search, randomasized search and Bayesian Optimization) and regression models (Polynomial Regression, Decision Tree Regression, Random Forest Regression, Support Vector Regression and Gradient Boosting Regression) for predicting unseen data. The results indicate that there is no universal method for predicting the parameters considered. For example, Random Forest Regression model with hyperparameter Grid Search method performed best, yielding an MSE of 0.0003 and R 2 of 0.96 for predicting TSS whereas Random Forest Regression model with hyperparameter Bayesian optimization method performed best for predicting COD resulting in a MSE of 0.0003 and R 2 of 0.97. Gradient Boosting Regression model with hyperparameter random search method performed best, yielding an MSE 0.0002 and R 2 of 0.90 for predicting BOD.