Reduction of input variables in the process of modeling risk possibility of elm trees by principal component analysis

Trees can cover a wide range of environmental, aesthetic, historical, tourism benefits and social, physiological, as well as economic benefits, however, the trees can cause environmental risks. The aim of current study was to reduce the insignificant variables by methods of index criteria of tree and analysis of the main components and then using the remaining variables in the process of modeling the risk possibility falling of elm tree by Neural Network Method. For this purpose, the quantitative and qualitative variables of elm trees were collected in Seyed Alikhan Street of Isfahan by counting 100% of these trees in 2018. The results of elimination of tree diameter and contact with power transmission lines in predictive models of risk of falling dangerous trees indicated that the variables that were eliminated during the study process did not have a significant effect on predicting the risk of falling dangerous trees. This creates a significant change in the accuracy of the model training data and decrease the number of repetitions of network training to reach the optimal network and reduces the time to reach the optimal neural network. Therefore, we recommended the use of tree fall modeling for management of urban green spaces.