Hedonic regression vs. neural network. A comparison in price prediction of residential apartment.

Abstract

Housing investment is always one of the most significant investments and requires the careful calculation and consideration. Therefore, the accuracy in prediction plays an important role in making investment decision. There are many types of housing price prediction and evaluating the property price based on that property’s constituted elements is one approach. This thesis attempts to predict the housing transaction price through the bundles of property components in Helsinki Metropolitan Area with two models: hedonic regression and neural network to answer the research question “Which model gives more accurate prediction: hedonic model or neural network model?”.

The empirical analysis also contains the potential complication if the model bias happens. That is the reason why the author would like to run the model through as many possibilities as possible. The dataset has been trained and tested for five times to find out the most accurate prediction as well as to avoid the model bias. The conclusion indicates that neural network delivers the more superior performance than hedonic regression by using R2 as the criteria to evaluate model performance. The best result of hedonic regression model 0.91, meanwhile the best result of neural network model is 0.95. It suggests that neural network can be a better alternative for housing price prediction, especially in the given dataset. This research’s result aims at consolidating the previous studies’ outcomes when these two models were compared with each other and neural network model also provided the better performance than hedonic regression.