Enhancing accuracy of extreme learning machine in predicting river flow using improved reptile search algorithm

This study searches the feasibility of a new hybrid extreme leaning machine tuned with improved reptile search algorithm (ELM-IRSA), in river flow modeling. The outcomes of the new method were compared with single ELM and hybrid ELM-based methods including ELM with salp swarm algorithm (SSA), ELM with equilibrium optimizer (EO) and ELM with reptile search algorithm (RSA). The methods were evaluated using different lagged inputs of temperature, precipitation and river flow data obtained from Upper Indus Basin located in Pakistan. Models performance evaluation was based on common statistics such as root mean square errors (RMSE), mean absolute errors, determination coefficient and Nash–Sutcliffe Efficiency. The prediction accuracy of single ELM model with respect to RMSE was improved by 2.8%, 7.7%, 15% and 20.7% using SSA, EO, RSA and IRSA metaheuristic algorithms in the test period, respectively. The ELM-IRSA model with lagged temperature and river flow inputs provided the best predictions with the RMSE improvement of 20.7%.