Optimization suspended load estimation models by using geo-morphometric parameters and attribute reduction technique

Estimation sediment load of rivers is the most important challenges in river engineering. So, it was addressed different models by varying structures to estimate sediment load. In this study, it was reviewed effectiveness of geo-morphometric parameters and data mining technique to predict suspended sediment load in 68 basins in two different regions of Iran. For this reason, it was run six artificial neural networks models, linear regression, K-nearest neighbor, Gaussian process, support vector machine evolutionary on two types of suspended sediment data (i.e. maximum and average sediment). To optimize models, it was used geo-morphometric parameters and river discharge as input data into model and it was used attribute reduction technique to decrease the algorithms space. Results of models evaluation indicated that models performance is difference in average and minimum sediment data, so that the best method to predict average sediment is the Gaussian model by correlation coefficient, 0.988 and root mean squared, 11.004 and the best method to predict minimum sediment is support vector machine evolutionary model by correlation coefficient, 0.966 and root mean squared, 0.171.