A Hybrid Machine Learning Model for Modeling Nitrate Concentration in Water Sources

• Published in: Water, air, and soil pollution Vol. 234; no. 11; p. 721
• Main Authors: Mazraeh, Adnan, Bagherifar, Meysam, Shabanlou, Saeid, Ekhlasmand, Reza
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.11.2023; Springer; Springer Nature B.V
• Subjects: air; Algorithms; Analysis; Aquatic resources; Atmospheric Protection/Air Quality Control/Air Pollution; Climate Change/Climate Change Impacts; Clusters; Contaminants; Earth and Environmental Science; Environment; Environmental monitoring; Hydrogeology; Kernel functions; Learning algorithms; Machine learning; Mathematical models; Mathematical optimization; Modelling; Monitoring methods; Nitrates; Optimization; Parameters; Pollution sources; Root-mean-square errors; soil; Soil Science & Conservation; Support vector machines; variance; water; Water pollution; Water Quality/Water Pollution; Water resources; Harris Hawk optimization (HHO); Nitrate; Sequential forward floating selection (SFFS); Jenks natural breaks algorithm (JNBA); Support vector machines (SVM); Machine learning
• ISSN: 0049-6979; 1573-2932
• DOI: 10.1007/s11270-023-06745-3

Nitrate is one of the most dangerous contaminants that can pollute water sources; as a result, it is always tried to use accurate methods to monitor its quantity. The goal of this study is to develop a hybrid machine learning model (HELM) for modeling nitrate concentration in water resources. For this purpose, 1453 samples were collected over a 20-year period (from 2000 to 2020). To develop the HMLM, the concentrations of nitrate data were first clustered using the Jenks natural breaks method (JNBA). After that, a support vector machines (SVM) model was developed for each cluster. In this case, the trial-and-error (TaE) method was used to determine the parameters of kernel functions. However, the sequential forward floating selection (SFFS) technique was employed to select the optimal input parameters to simulate the nitrate content in each cluster. In the last step, to improve the efficiency of the SVM model, the parameters of the kernel functions were determined using the Harris-Hawkes optimization (HHO) algorithm. The HHO algorithm is used because of its ability to intelligently shift between the exploration and exploitation phases during optimization. To develop all of the models, 80% of the data was utilized for training and the remaining for testing. Finally, indices such as root mean square error (RMSE), mean absolute error (MAE), Willmott’s index agreement (WI), explained variance score (EVS), Kling-Gupta efficiency (KGE), and coefficient of determination ( R 2 ) were used to evaluate different models. Based on the results of this study, HMLM can be used to accurately model nitrate concentration. Furthermore, using the HHO optimization algorithm instead of the TaE method considerably improves the SVM model’s performance.