Assessment of the Effect of Input Factors Number in Accuracy of Artificial Neural Network for Landslide Hazard Zonation (Case study: Haraz Watershed)

More than 30% of Iran's land is formed from mountainous areas. So each year, landslides cause damages to structures, residential areas and forests, creating sedimentation, muddy floods and finally deposit the sediments in reservoir dams. Therefore, for preventing of this damages and expressing the sensitivity rate of hillslopes, landslide hazard zonation is considered in prone areas. The purpose of this study is to determine the optimal structure of artificial neural network with different numbers of input factors for the landslide hazard zonation in the Haraz Watershed. First, the number of optimal epochs was determined to prevent network overlearning with trial and error method. Then, 14 neurons were determined in the hidden layer. Finally, the number of neurons was changed from 1 to 9 in the input layer. According to the obtained results, with increasing the number of neurons in the input layer, efficiency of Artificial Neural Network improved for landslide susceptibility mapping. In this research, nine neurons in the input layer, 14 neurons in the hidden layer and one neuron in the output layer were selected as the optimal structure. Root Mean Square Error and Descriptive Coefficient (R2) were equal to 0.051 and 0.962, respectively and the accuracy of landslide hazard zonation map was equal to 92.3%. Meanwhile, the results showed that about 35.14, 26.73, 14.59, 9.88, and 13.63 percent of all studied areas are located in stable, low, moderate, high and extremely hazardous areas, respectively.