Performance Evaluation of Ensemble Learning Techniques for Prediction

The paper employed a range of ensemble learning techniques to forecast Chemical Oxygen Demand (COD) levels in Wastewater Treatment Plants (WWTP). By harnessing ensemble learning, the main objective is to enhance the accuracy and robustness of COD prediction models by capitalizing on the diverse strengths of multiple algorithms. Our experiment encompassed two distinct WWTP datasets, allowing us to comprehensively assess the efficiency and effectiveness of three ensemble methods. Firstly, Bagging operates by creating multiple versions of a predictor and combining their outputs to form an aggregated prediction. Secondly, Boosting algorithms such as AdaBoost, GBM, and XGBoost sequentially constructed powerful learners, focusing on previously misclassified instances to bolster predictive performance. Finally, Stacking methodologies combined predictions from multiple models using a meta-learner. The analysis of the results for various models across the two wastewater treatment plant datasets (WWTPDS1 and WWTPDS2) reveals distinct performance levels. The Bagging model exhibits a reasonable performance with WWTPDS1 (MSE: 0.0016, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>: 0.9351) and a slightly lower performance with WWTPDS2 (MSE: 0.0033, R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>: 0.6787). The Boosting model outperforms with lower MSE and higher $\mathrm{R}^{\mathbf{2}}$ values for both datasets (WWTPDS1 MSE: 0.0010, $\mathbf{R}^{2}$: 0.9751; WWTPDS2 MSE: 0.0028, $\mathrm{R}^{2}: 0.7384$), indicating its better predictive accuracy and reliability. The Stacking model evaluated in the training phase, it showcases exceptionally low MSE and high $\mathrm{R}^{\mathbf{2}}$ values (WWTPDS1 MSE: 0.0004, $\mathrm{R}^{\mathbf{2}}$: $\mathbf{0. 9 7 1 2;}$ WWTPDS2 MSE: $0.0009, R^{2}: 0.9213$). In the testing phase, the Stacking model maintains strong performance (WWTPDS1 MSE: 0.0013, $R^{2}: \mathbf{0. 9 4 9 4}$; WWTPDS2 MSE: 0.0028, $\mathrm{R}^{2}: \mathbf{0. 7 1 9 1)}$.