Housing Rent Prediction Model Based On Ensemble Learning

Housing is closely related to people’s wellbeing.In the current social environment,the unbalanced economic development leads to the continuous accumulation of population in first-and second-tier cities,which brings more demand for rental housing,and renting has gradually become an important way of people’s daily living,the rapid growth of its demand has made the rental market flourish.Rent is a key factor for the stable and healthy development of the rental market.The prediction and analysis of future housing rent can give a certain trend warning to tenants,landlords and rental agencies,and provide reliable technical support for the government to regulate the rent level.However,there is a lack of research on rent forecasting,and most of the employed models still have the problem of insufficient accuracy.Therefore,this thesis builds a rent prediction model based on ensemble learning theory,aiming to accurately predict future housing rents and provide reliable data reference for multi-party entities in the rental market and government departments.Based on the hedonic price theory,this thesis divides the factors affecting housing rent into four categories:building,location,neighborhood and other features.Firstly,observing the distribution characteristics of each variable through descriptive statistical analysis to understand the outliers and missing values of the data;Then,data cleaning and feature coding are carried out for the chaotic original data;Next,using the mutual information method to filter the features to further improve the data quality;Finally,rent data from January to June are sampled to obtain the training set,and the data from July is used as the test set,four single models of Random Forest,XGBoost,Light GBM and Cat Boost were constructed.At the same time,two methods of bayesian optimization and particle swarm optimization were used to adjust the important parameters of the model,and RMSE,MAE,and R~2were selected as model evaluation indicators.The experimental results show that in terms of parameter optimization,the parameter tuning effect of bayesian optimization is significantly better than that of particle swarm optimization.After bayesian optimization tuning,the RMSE,MAE and R~2of Cat Boost model are 0.1623,0.2412 and 0.8935,respectively,which are better than other models in the three evaluation indicators.The results show that the Cat Boost model based on Bayesian optimization algorithm has higher prediction accuracy,and has better prediction effect for rent prediction.