Modeling stream dissolved oxygen concentration using teaching–learning based optimization algorithm

The aim of this study is to investigate the applicability of teaching–learning based optimization (TLBO) algorithm for the first time in modeling stream dissolved oxygen (DO) prediction. The input parameters selected from a surface water-quality study including 20 indicators for the models are water pH, temperature, electrical conductivity, and hardness, which were measured semimonthly at six monitoring sites selected in an untreated wastewater impacted urban stream during a year, due to their direct and indirect effect on DO concentration. The accuracy of TLBO method is compared with those of the artificial bee colony algorithm and conventional regression analysis methods. These methods are applied to four different regression forms: quadratic, exponential, linear, and power. There are 144 data for each water-quality indicator, 114 of which are designated for training and the rest for testing patterns in the models. To evaluate the performance of the models, five statistical indices, i.e., sum square error, root mean square error, mean absolute error, average relative error, and determination coefficient, are used. The TLBO method with quadratic form from among all models yielded better prediction, with an improvement of nearly 20 %. It can be concluded that the equations obtained by employing the TLBO algorithms predict the stream DO concentration successfully. Therefore, the employment of the TLBO algorithm by water resources and environment managers is encouraged and recommended for future studies.