<![CDATA[Purpose: This study aims to improve the accuracy of water quality classification models by addressing class imbalance and data noise. Aquaculture water quality monitoring is essential to support sustainable aquaculture production and maintain aquatic organism health. As aquaculture systems become more intensive, accurate predictive models are needed for effective monitoring and decision-making. Therefore, this study proposes a stacking ensemble model optimized using SMOTEENN hybrid balancing to improve classification performance on aquaculture water quality datasets. Methods: The proposed approach combines SMOTEENN hybrid balancing with a meta stacking ensemble framework. First, Synthetic Minority Oversampling Technique (SMOTE) was applied to balance minority classes by generating synthetic samples. Next, Edited Nearest Neighbor (ENN) was used to remove noisy data from both original and synthetic datasets. After preprocessing, the classification process employed a meta stacking ensemble model consisting of Extra Trees (ET) and Random Forest (RF) as base learners, while Logistic Regression (LR) served as the meta learner. The model was evaluated using the Aquaculture Water Quality (AWQ) dataset. Result: Experimental results show that the proposed model achieves the highest performance under the SMOTEENN scenario, reaching an accuracy of 99.88%, outperforming SMOTE 99.20%, ENN 99.78%, and the original imbalanced data 99.18%. The results indicate that combining class balancing and noise reduction significantly improves classification performance. Novelty: This study presents a novel integration of SMOTEENN hybrid balancing and meta stacking ensemble learning, offering an effective solution for handling imbalanced and noisy environmental datasets in water quality classification]]>
