An Evaluation of Gene Expression Programming and M5 Models in Daily Discharge Prediction: A Case Study of Lighvan River

Increasing water consumption per capita in one hand and evident climate changes on the other hand، will result in shortage of water resources in the near future. In recent decades، various methods based on artificial intelligence such as artificial neural networks، fuzzy logic، neuro-fuzzy and genetic programming have been developed to forecast and create models of different parameters in water engineering. In this study، the daily river flow data of Lighvan Basin in East Azarbaijan Province was used to model the flow، by using Gene Expression Programming and M5 methods with 5 different flow delay models. According to the results of the models، for the daily flow the best model in Gene Expression Programming method with R2=0. 929 and RMSE=0. 212 and the best model in M5 method with R2=0. 926 and RMSE=0. 216 obtained. The results of comparing statistical indicators، RMSE، MSE، and R2، showed that both Gene Expression Programming and M5 models have acceptable accuracy in daily discharge prediction. Although M5 model has less mean bias error in comparison with Gene Expression Programming، overall it is obvious that Gene Expression Programming is relatively better than M5 in river flow prediction.