Applicability of artificial neural network in predicting house rent

Abstract

House rent prediction has great importance in real estate development as wel! as ill overall housing situation of a city, The various participants in the real eslate market have a substantial interest in the prediction of house rent. Rent models can be an effective tool when empirical data carum! be collected either because of practical constraints of cost, time etc. or when future scenarios are being dealt with. Hedonic price (multiple regression) models have been commonly used to estimate house rent. To address the issue of application of Artificial Neural Network (ANN) in house nonl prediction, thIS study aims to develop an artificial neural network model for house rent prediction. The study will also use the results from a hedonic price model for house rent prediction and compare the predictive power of both models. The data setu.sed to develop the Neural Network Model cOllsists of a sample of 479 single family and multi-family residential properties available for rent in Rajshahi City. The neural network model built for this data set utilized fourteen independent variables. The neural network models developed in this study arc the "beSI" models that were obtained utilizing a sequential trial and error method. The best model developed with eighty hidden neurons had the Rl valu.e of 0.621 for sample forecast. The study has demonstrated that neighborhood attributes arc the most significant factors in detemlining the house rent of Rajshahi City. The percentage of area dedicated to community facilities and percentage of area dedicated to eommereial use have contributed more to the predictive power of model than the other attribntes. So it is seen that land usc has a great impact on honse rent in Rajshahi City. The study also empirically compares the predictive power of the artificial neural network lllodel with the hedonic price model on house rent prediction. The eomp!,rison was conducted in six stages or cases. The results indicate that the neural network model ou.tperformed the hedonic price model in all of the cases. In th,S study, the ANN model consistently gave better result than the hedonic price model, although the difference between the two models was not too large. ANN model and hedonic price model both do better when they are trained and tested with the same data set but they perfonned poorer on out-of -sample forecast. But in both cases ANN model showed better results in comparison to hedonic price modeL The study also supports the superiority of ANN model in prediction of outlier holdout sample.