artificial neural network

In this applied study, analysis of Iranian society and its threshold of tolerance has been done in a descriptive-analytical method. The society were professors of universities and the sampling size were 133. Data collection was done in archive and field methods. The reliability of the questionnaire was 89%. A multilayer perceptron artificial neural network was created to predict and model the human development index of the 38 target countries. Four input parameters were life expectancy at birth, expected years of schooling, mean years of schooling, and gross national income per capita". The response parameter was the human development index. The 4000 structures were developed, trained, validated, and tested by changing the variables of activation function, training function, and the number of neural neurons in the hidden layer. The optimal neural network in terms of network performance measures was the network with mean absolute percentage error (MAPE = 0.01156), Normalized Root Mean Square Error (NRMSE = 0.00124), and R squared (R2=0.99620). The optimal network had ten neurons in the hidden layer, a tansig activation function, and an RP training function. The artificial neural network model had a high accuracy in predicting the human development index. Based on the results of sensitivity analysis, the most influential factors were mean years of schooling, "life expectancy at birth, expected years of schooling, and gross national income per capita, with the relative importance of 33.63%, 24.43%, 23.83%, and 18.11%, respectively.

Modeling the national power using explicit mathematical equations is very difficult. Artificial neural networks can solve complex multi-dimensional problems. In this applied study, The population was professors of the military, social sciences, economics, and geopolitics specialties. The sampling method and the sample size were randomly and 133, respectively. Data collection was done in archive and field methods. The reliability of the questionnaire was 89%. The obtained data were modeled by a multilayer perceptron neural network with nine input parameters including the dimensions of national power, and a response parameter i.e., the national power of the target countries. The 8000 neural network structures were developed by changing the variables of activation function, training function, and the number of neural neurons in the hidden layer. The results showed that the best neural network characterizations were the tansig activation function, LM training function, and 16 neurons in the hidden layer. The efficiency values of the network were MAPE = 0.00071, NRMSE = 0.00006, R2= 1.00000 in the training stage, MAPE = 0.00579, NRMSE = 0.00137, R2= 0.99999 in the test stage and MAPE = 0.00300, NRMSE = 0.00124, R2= 0.99997 in the validation stage. The relative importance of the national power dimensions were military 15%, scientific 13%, territorial 13%, cross-border 12%, social 11%, economic 10%, spatial 10%, political 8%, and cultural 8%. Based on the results, the optimal neural network model could predict the national power of the target countries with very high accuracy.