Housing Rent Prediction Based On Stacking Fusion Model

The rental market is a very important part of China’s housing market.Due to the high price of commercial housing in recent years,more and more people prefer to rent a life way to solve the housing problem.With the rapid development of Internet technology,such as the rapid rise of online housing rental platform appeared in people’s vision,its convenience and efficiency are deeply loved by people.But in huge rental market,there are a series of problems,such as the landlords cannot accurately carry on the reasonable pricing of housing rental,and asymmetric information between landlords and tenants.These problems have been resulted in the waste of housing resources to some degree,and have been hindering the development of rental market.Therefore,it is especially important in the price research on housing rent.This paper does the following research based on the existing problems in huge rental market.First,we get the real rental data of Shenzhen city on Lianjia from January 2021 to May 2021 using Web crawler technology,and crawl the basic supporting data of the surrounding housing sources based on Baidu API platform in order to expanding the range of characteristic variables of the data.Then,the original data set was preprocessed.From descriptive statistical analysis of data,we preliminarily explore the relationship between each characteristic variable and the target variable rent.In order to facilitate the construction of the later forecasting model,we carried out data transformation work such as Discrete variable unique heat coding,continuous variable normalization.Second,we built a single model of housing rent prediction based on random forest,XGBoost and Light GBM to train data.In order to improve the accuracy of prediction,we use 10-fold cross validation when train the models,and optimize the important parameters of each model by using the grid search method.Besides,the mean absolute error,the root mean square error and the coefficient of determination are used as indices to evaluate the performance of a prediction model.By comparing the evaluation criterion results of the three single machine learning models,it is found that the Light GBM prediction model has smaller error and larger goodness of fit value after adjustment,which shows that the Light GBM model is more suitable for rental prediction in this experiment.Final,three single models with certain prediction,Random Forest prediction model,XGBoost prediction model,Light GBM prediction model were used as base learners and use MLR as the meta-model by the Stacking methods to construct fusion prediction models.The final experimental results show that the Stacking fusion model performs better than the other three single base prediction model on the data set,and the R-square of the Stacking fusion model can reach 0.932,the mean absolute error is0.075,and the root mean square error is 0.123.It suggests that multi-model fusion makes full use of the advantages of each single model,and improves the prediction effect and generalization performance of the model.