nitrate adsorption modeling using svm and lssvm models

Objective: nitrate compounds are among the pollutants of groundwater resources, and in recent years, in terms of agricultural development and human activities, their average rate has been increasing. the purpose of this study is modeling of nitrate adsorption by using svm, lssvm and random forest model. material and methods: the nitrate adsorption data used in this study were first randomized and standardized and then divided into two groups of training and testing. 70% of the data were in the training group and the remaining 30% in the experimental group. validation of model training was performed using k-fold cross-validation method with a value of k = 5 in order to prevent over-fitting of models. the parameters of random forest, svm and ls-svm models were determined using a bayesian optimization algorithm. the objective function of the optimization algorithm was to minimize the mse error value of the model.results and discussion: based on the results, the random forest model was used with the bagging algorithm and the parameters of minimum node size, number of trees and number of variables used were equal to 2, 10 and 3, respectively. the svm model was trained with the rbf kernel function and the parameters of box constraint and epsilon equal to 2.2156 and 0.0891, respectively, along with standardization of input and output data of the model. the ls-svm model was also trained with rbf kernel function and setting parameters and kernel function equal to 3160/3160 and 19.7891/19, respectively. conclusions: taylor diagram results showed that the stochastic forest model and svm had a higher correlation between observational and estimated data. therefore, based on the results, the stochastic forest model is more consistent with the observation data and predicts nitrate changes well.