Building an Efficient Algorithm for Long Bien District and Montreal Real Estate Pricing

Abstract

   The LASSO method regularizes linear regression coefficients by adding a ℓ₁ norm penalty to the least square criterion. Recently, this method has been used very popularly to solve high dimensional regression problems in statistics, data mining, and machine learning for big data. In this paper, we applied the LASSO method to regularize nonlinear regression coefficients for the real estate pricing problem. Real estate pricing was often based on a few dozen features, and obviously the relationship between real estate prices and their features is nonlinear. Therefore in the present study we used a nonlinear model and applied LASSO method to regularize the coefficients. Because the performance of LASSO application is sensitive with regularization parameter, we proposed an aggregation of LASSO nonlinear regression combining weak LASSO regressions to produce a robust one which has smaller variance. This algorithm was evaluated on the real estate datasets collected in Montreal province, Canada (Noseworthy, 2014) and in Long Bien district of Hanoi and more accurate results than the state of the art algorithms were obtained.