Prediction of water quality parameters using support vector regression

Efficient water management and quality quantification can be greatly aided by more precise prediction of water quality parameters. The most important water quality parameters for analysing effluent or river water are the biochemical oxygen demand (BOD) and chemical oxygen demand (COD), since they show the level of organic matter. Standard approaches like Winkler-Azide and the reflux method are used to compute BOD and COD for any water sample; however, these methods are time consuming and prone to measurement error. Consequently, data-driven technique (DDT) approaches can be used here along with traditional techniques. In the present research, a kernel-based support vector regression (SVR) technique is utilised for predicting BOD and COD for three distinct sections of the Mula-Mutha River in Pune, India. Kernel-based approaches provide a powerful framework for motivating algorithms that act on general types of data and provide general relations such as classifications and regressions. The Pearson VII universal kernel (PUK kernel), the radial basis function (RBF) kernel, poly kernel (POLY) and normalised poly kernels (NPOLY) were used in the current study. The proposed models were assessed using the coefficient of correlation (R), root-mean-square error (RMSE), and mean absolute relative error (MARE). Overall, the PUK-based model outperformed the RBF and poly kernel-based model in terms of accuracy, with a lower RMSE and higher r. Visual assessment of models developed using kernels, i.e. scatter plots, box plots and Taylor diagram, shows that models developed using PUK and RBF are better predictors than other models.