Almost most of the installations located in natural beds face many threats over time, and in order to reduce the damage, it is necessary to identify the threatening factors and use the results in the appropriate location or in taking measures to reduce the damage. Today, the increase in consumption Gas has caused an increase in the density of the gas transmission pipeline network and, as a result, an increase in its potential risks. The first step in risk analysis is to identify the effective factors in the occurrence of accidents and breakdowns on the pipeline. According to the environment around the pipeline, various factors cause pipeline damage and accidents. After identifying the damage factors, the amount of damage caused by each is calculated and the results are expressed in the form of risk. It is possible to establish a connection between the risk estimation process and the geographic information system and make appropriate zoning. It was determined by using models and geographic information system. Therefore, in this research, the process of estimating environmental risk with geographic information system and using hybrid-fuzzy algorithms has been investigated. Valuable information such as risky components can be determined by assessing the risk of gas pipelines, and a suitable response and strategy can be used to reduce or even eliminate it. In order to achieve this goal, it is necessary to use a suitable technique that can accurately and reliably assess the existing risks, so that planners and managers can act with a wider horizon and a lower risk factor towards the optimal management of gas transmission lines. The extent of gas lines in Tehran and Qom province and considering the environmental and natural characteristics of the two provinces, it is very important to assess the amount of damage. In this research, MATLAB version 2019b software was used in order to assess the risk of the gas pipeline using the multi-layer perceptron neural network model. Due to the fact that the number of input nodes and hidden layers are varied in the specified range, the optimal number of input nodes and hidden layers was determined by model selection criteria on the test data and the WIC model was used. Due to the existence of 11 criteria in this research, 11*11 modes were created. Also, in this research, the input layer has 6 neurons and 1 neuron in the hidden layer and the algorithm used is Levenberg- Morquardt according to the purpose of the research and high accuracy. In this model, there were 740 data, 70% of which were used for training, 15% for testing, and 15% for evaluation. For risk assessment, criteria were weighted by VIA method. In this method, using the feature eliminate process, a criterion was removed in each step and the network error was measured. In this research, in addition to the MLP model, the Random Forest model was used. In this method, an estimate of the classification error can be obtained based on the training data. The number of trees should be enough to stabilize the error rate and the additional index that is created in the RF method. To estimate the feature importance, first the OOB components are run among the trees and the votes are counted for correct classification. Then, the prediction accuracy is obtained many times after randomly changing all the values of this feature while all other features are the same. In order to assess the risk of the gas pipeline using the random forest model from MATLAB version 2019b software and from the model Regression and RF Regression function were used. In this model, the selection of training and test data is random and 740 points are samples, 80% of which include training data and 20% of test data. In this model, the number of decision trees used by the tree bagger function is 500. To check the validity of the models, the estimated values obtained from the networks and the measured values in the test phase were used. To validate the model, the root mean square error (RMSE), mean error of exploitation (MBE), and mean absolute error (MAE) were used.According to the results of ANP landslide index, 0% of the area is in the low risk class, 17.28% in the medium risk class, 73.14% in the high risk class, and 9.58% in the high risk class. Therefore, it can be said that 19.008 km of the investigated area are located in parts with moderate vulnerability, 80.454 km with relatively high vulnerability and 10.538 km with high vulnerability. According to the results of the Fuzzy model landslide index, 41.85% of the area is in the low risk class, 11.60% in the medium risk class, 22.52% in the high risk class, and 24.03% in the high risk class. Based on the landslide criterion and the results of the fuzzy model, it can be said that 46.035 km with low vulnerability, 12.76 km with moderate vulnerability, 24.772 km with relatively high vulnerability and 26.433 km with vulnerability. In this research, the systematic error (MBE) of the MLP model is estimated to be 0.002812, and the absolute error of the model is 0.042168. The RMSE error rate is 0.05020. The systematic error (MBE) of the RF model is -0.151848. The absolute error of the model is 0.179101. The systematic error (MBE) of Fuzzy_ANP model is -0.16893. The absolute error of the model is 0.170337. The RMSE error was 0.12262.