On Prediction Of Boutique House Price Based On Regression Ensembles

Recently,with the rapid development of China's economy,the real estate in-dustry has risen rapidly.With the accelerated pace of people's life,hardcover houses are becoming more and more popular.So it has important guiding significance to quickly and accurately predict multi-variable hardcover house to buyers or housing sellers.In this paper,the hardcover room in Ames area is taken as the research object due to the lack of hardcover room data in China.Firstly,the correlation-based in-terpolation of regression model and the categorical-variables-based mode or k-means interpolation are used to fill in the missing values.Secondly,according to the record-ed characteristic variables,new influential variables are constructed.The data are also being logarithmically transformed and normalized.Then some single prediction modes are established based on the training set,such as linear regression,Ridge re-gression,Lasso regression,etc.The results show that the Lasso regression prediction is the best,which has the root mean square error=150578,~2=0.926347.Since the single model prediction of each sample point is not consistent,a weight combination model as well as a stacking combination model is established.In the weight combination model,prediction variances of single models,Ridge regression model,Lasso regression model,ElasticNet regression,Gradient Boosting regres-sion,XGBoost regression,Random forest regression,are converted into weights for weighted combination.Compared to Lasso regression model,RMSE on test set re-duced by 6.9%and~2increased by 1%.In the stacking combination model,the predictions of six regression models were considered as new feature variables to es-tablish a linear regression model.Compared to the Lasso regression model and the weight combination model,the RMSE of the stacking integration model is reduced by 22.3%and 16.6%respectively,~2is increased by 1%and 2.3%respectively,in-dicating that the stacking integration algorithm has the most accurate prediction effect and low error on Ames'house price,which can be a good reference value for buyers or sellers.