An annealing dynamical learning-based relevance vector regression algorithm for housing price forecasting

Accurate housing price prediction is very important to make the national macro-control policy. In order to solve the quadratic programming problem of traditional relevance vector regression, an iannealing dynamical learning-based relevance vector regression algorithm (ADLRVR) is first presented to predict housing price in the paper. Housing price data of Shanghai city from 2011-5-29 to 2011-6-27 are employed to testify the prediction performance for housing price of the proposed ADLRVR method. The ADLRVR models with the 3~7 input nodes respectively are trained to find the optimal number of input nodes of the prediction models due to the great influence on the prediction effects of the number of input nodes of the prediction models. The prediction results show that the ADLRVR model, the RVR model and the SVR model with 5 input nodes have the best effects, which are 0.48%, 1.15%, 3.0% respectively. It is indicated that the prediction results for housing price of ADLRVR are better than those of the RVR model and the SVR model. 1548-7741/