Classification of water quality status based on minimum quality parameters: application of machine learning techniques

• Published in: Modeling earth systems and environment Vol. 4; no. 1; pp. 311 - 324
• Main Authors: Dezfooli, Donya, Hosseini-Moghari, Seyed-Mohammad, Ebrahimi, Kumars, Araghinejad, Shahab
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.04.2018; Springer Nature B.V
• Subjects: Accuracy; Case studies; Chemistry and Earth Sciences; Classification; Coliforms; Computer Science; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Ecosystems; Environment; Errors; Learning algorithms; Machine learning; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Mathematical models; Neural networks; Original Article; Parameters; Performance assessment; Phosphates; Physics; Removal; Rivers; Statistics for Engineering; Support vector machines; Turbidity; Water analysis; Water quality; Water sampling; Probabilistic neural network; NSFWQI; Support vector machine; Classification; K-nearest neighbor
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-017-0406-9

This paper focuses on three models namely probabilistic neural network (PNN), k-nearest neighbor and support vector machine (SVM) as an alternative to NSFWQI in order to classify water quality of Karoon River, Iran as a case study, regarding minimum possible parameters. For this purpose a set of 172 water samples were used in a way that water quality parameters and their water quality classes (from NSFWQI) were considered as the input–output of the models, respectively. Three assessment criteria, namely error rate, error value and accuracy, were applied in order to assess the performance of the applied models. The results revealed that under the condition that no parameter is removed, all the three models showed the same results. However, under quality parameters removal the results revealed that PNN was the best model, as that managed to classify water quality solely based on three quality parameters of turbidity, fecal coliform and total solids, at 90.70% accuracy, 9.30% error rate and error value was 4. In contrast to PNN, in the same condition, SVM showed the poorest performance. As with making use of four quality parameters namely fecal coliform, DO, BOD and total phosphates, it classified water quality at 88.37% accuracy and 11.63% error rate.