Enhanced Aquaculture by Water Quality Assessment Using Discriminant Analysis Algorithm

Water quality is essential for sustainable aquaculture, but its variability makes assessment difficult. This study compares linear and quadratic discriminant analysis for classifying aquaculture water conditions using a public dataset of 4,300 observations with 14 physicochemical and biological parameters. Data preprocessing included outlier removal and feature standardization, followed by a stratified train-test split without hyperparameter tuning. The quadratic discriminant analysis consistently outperformed the linear model, achieving 93.24% accuracy, 93.75% precision, and 92.58% F1-score—approximately 5.8% higher than the linear variant. Notably, for the critical poor water quality class, the quadratic model delivered perfect precision and significantly better recall, along with near-ideal area under the precision-recall curve, ensuring reliable early warnings without false alarms. In contrast, the linear model, restricted by the assumption of equal covariance matrices, failed to capture the heterogeneous variance in environmental data, resulting in unacceptable false negatives. These findings establish the quadratic approach as a statistically sound and interpretable alternative to complex black-box models, emphasizing the need to match algorithmic assumptions with data complexity in aquaculture monitoring.