پیش‌بینی قیمت مسکن برای شهر اهواز: مقایسه مدل هدانیک با مدل شبکه عصبی مصنوعی

Determination and the estimation of the house price in urban areas has a great importance for governments, individual and state investors and common people. The mentioned estimation can be used in future planning and decision making of many urban and regional policies. In this regard, due to the vital importance of the house price in recent decades powerful and effective functions have been used in order to forecast and estimate the house price. The main purpose of the current study was to present an optimal model to forecast and identify the determinants of the house price in Ahvaz metropolitan by the comparison of the Hedonic and Artificial Neural Network Models and the choice of the optimal model. The present study has a developmental-applied nature which follows a descriptive-analytical approach. The sample included 286 statistical unit in 2015 based on 27 related variables were evaluated to forecast the house price in Ahvaz city. This study used semi-log hedonic function and Neural Network Multilayer Perceptron (MLP) approaches. To compare the results of the two models in terms of predictability the criteria of R2, MSE, RMSE, MAPE, MAE and TIC coefficient were utilized. The results of the Hedonic model indicated that that among 27 variables, 18 variables were significant model and by the comparison of the results and the estimated value, it turned out that housing prices in Ahvaz is mainly influenced by the physical and structural factors. Moreover, the comparison between all criteria demonstrated that the Artificial Neural Network had a better performance than the Hedonic regression model in forecasting the house price. In order to test the difference of precision in forecasts of the house price in Ahvaz the Morgan-Granger-Newbold test was conducted as an appropriate instrument. The results of test indicated that the there was a significant difference between the predicative power of the two models which confirmed the greater performance of the ANN (%98) model in comparison with Hedonic regression model (88%).