Research and Demonstration of a Prediction Algorithm Based on GBDT-LightGBM Algorithm

Abstract

This paper aims to use the LightGBM algorithm to predict housing prices in Chinese municipalities. According to previous research experience, from the demand level, supply level, and regulation policy three main aspects as the main influencing factors of housing prices are analyzed and predicted. In the case analysis, the determination coefficient (R-Square) and the average absolute percentage error (MAPE) are used to test the accuracy of the model, and the Kendall tau-b (K) method in the Kendall coefficient is used for correlation analysis and consistency test. After eliminating the repeatability index, the Kendall coordination coefficient W of the model is 0.977. After selecting the appropriate influencing factors and data, the Light Gradient Boosting Machine is trained by using the gradient boosting decision tree GBDT as the base learner and SPSS-PRO software. It is found that the model has the highest accuracy when the number of base learners is 500. From the training results, the influence degree of demand and policy level is the largest, and the influence degree of supply level is small. The influence degree of the three is 47 %, 43 %, and 10 %. In the secondary indicators, the main business tax and additional, urbanization rate, and a loan amount of real estate development enterprises have a greater impact. The R-square of the training set and the test set are 0.905 and 0.902, respectively. The accuracy of the model is high, which provides an effective reference for housing price prediction.