Comparative evaluation of meta-heuristic algorithms for hyperparameter optimization in ML-based water quality prediction

D Mu, SADHANA-ACADEMY PROCEEDINGS IN ENGINEERING SCIENCES, 50, 313 (2025).

DOI: 10.1007/s12046-025-02959-9

Providing safe and sanitary water is considered one of the most essential human challenges in societies, especially developing ones. This exploration offers a hybrid classification model drawing on machine learning (ML) to predict water quality. The dependent variable is a binary variable that predicts whether water bodies are potable, based on nine different water quality metrics. Due to the significant number of missing values in the database, two different approaches were used to prepare the data, including dropping and filling data. The main algorithms used for data classification are based on ML, including Extremely Randomized Trees (Extra Trees) and Extreme gradient boosting (XGBoost) Classifiers. Also, to boost the accuracy of predictions, seven meta-heuristic algorithms were used to optimize and adjust the hyperparameters of the main algorithms. Drawing on the findings of this exploration, it is recommended to use the Extra Trees Classifier in combination with the Manta Ray Foraging Optimization Algorithm (MRFO) optimizer, using the dropped database, to predict water quality. Additionally, when using the filled database, the Sparrow Search Algorithm (SSA) optimizer provides the best evaluation index values for optimizing the hyperparameters of the Extra Trees Classifier. These outcomes suggest a practical, efficient approach for enhancing automated water quality assessment, potentially supporting faster and more reliable monitoring efforts compared to traditional methods.

Return to Publications page