COMPARING THE NEURAL NETWORK WITH PATH ANALYSIS IN FITTING REGRESSION MODELS

Abstract. The purpose of this study was to compare a neural network with the path analysis in fitting regression models. The conceptual model of path analysis according to the studied data includes a dependent variable, two independent variables and one mediating variable. The neural network conceptual model is considered with three layers (input, hidden, output) and the hidden layers have two nodes. The study asked 474 people about their education, beginning salary, previous experience and their current salaries. The data divided into the train and test groups at the rate of %60 and %40. The criterion for comparing the two methods is RMSE. The results of the analysis showed that both models are over fitted and the RMSE train and test of neural network are less different from the path analysis. Therefore, in this dataset, it can be said that the neural network performs better than the path analysis.