Fine-Grained Real Estate Estimation Based On Mixture Models

People always want to know how much does the apartment they select really value, which is more useful than the whole trend of the housing price in one area for buyers. However, the transaction records about one certain kind of apartments are few or often lacked. In this paper, a novel estimation concept, fine-grained real estate estimation, is proposed and aims at evaluating the price of an apartment in one garden. One problem of the fine-grained estimation is that data is sparse. To deal with this sparsity problem, the strategy of applying the mixture model is put forward to alleviating the data sparsity. The experiments are all conducted in the real data from six districts second-hand housing transaction records (Between January 1st and July 15th, 2015) in Beijing, China. We find that proposed mixture model method has double better accuracy compared with other five traditional approaches in MAE and RMSE metrics.