Faculty of New Sciences and Technologies, University of TehranIranian journal of Ecohydrology2423-60985120180321Select the Most Suitable Inputs to the Artificial Neural Network Model by Using the ACO AlgorithmSelect the Most Suitable Inputs to the Artificial Neural Network Model by Using the ACO Algorithm59686482410.22059/ije.2017.230717.538FAMohammad Javad ZeinaliAli ShahidiAssociate Professor Department of Water Engineering College of Agriculture University of BirjandJournal Article20170408the performance criteria have used in this study is including mean square error (MSE), sum square error (SSE), Nash_Sutcliffe and correlation coefficient. The result indicated the best ANNa model is ANNa2 with MSE equal 0.0017. Inputs in this model are Total Cation, Total Hardness and Calcium. The best ANNb model is ANNb3 with MSE equal 0.0012. Inputs in this model are Sodium adsorption ratio, pH, Total Hardness and Calcium. Also, the results indicated that using ACO algorithm for finding the best input parameters had increased neural network performance, in ANNb models than ANNa for validation network and for test network we see with increases inputs the performance of network increases. According to results we can say that against try and error for finding the best inputs, we can use the parameter that those had a high correlation between target parameter as first step. But parameters that have high correlation between target parameter, necessarily don't the best inputs. But the parameter that those had a high correlation between target parameter as inputs of neural network. Also, we find that the ACO algorithm can be used as a method of input variable selection and that improved the performance of neural network.the performance criteria have used in this study is including mean square error (MSE), sum square error (SSE), Nash_Sutcliffe and correlation coefficient. The result indicated the best ANNa model is ANNa2 with MSE equal 0.0017. Inputs in this model are Total Cation, Total Hardness and Calcium. The best ANNb model is ANNb3 with MSE equal 0.0012. Inputs in this model are Sodium adsorption ratio, pH, Total Hardness and Calcium. Also, the results indicated that using ACO algorithm for finding the best input parameters had increased neural network performance, in ANNb models than ANNa for validation network and for test network we see with increases inputs the performance of network increases. According to results we can say that against try and error for finding the best inputs, we can use the parameter that those had a high correlation between target parameter as first step. But parameters that have high correlation between target parameter, necessarily don't the best inputs. But the parameter that those had a high correlation between target parameter as inputs of neural network. Also, we find that the ACO algorithm can be used as a method of input variable selection and that improved the performance of neural network.https://ije.ut.ac.ir/article_64824_69b3a7a4a209f29819d858eede62dcc5.pdf