Comparison of Real Estate Price Prediction Based on LSTM and LGBM

Abstract

Housing price prediction is one of the widely discussed topics, and the application of machine learning and deep learning models to housing price prediction is a research hotspot. Exploring which models are suitable for predicting real housing price datasets has significant implications for guiding government and homebuyers' decision-making. In this study, the LSTM and LightGBM models were selected as research objects, and the suitability of the models was explored and compared using the second-hand housing price dataset in Beijing. Based on the analysis, in the task of pure time series prediction based on historical housing prices, the LSTM model had a better fit (R2=0.91), while in the task of housing price prediction based on multiple influencing factors, the LightGBM model had better comprehensive evaluation index results (R2=0.53). Both models can be used to predict pure time series, but the LSTM model is not suitable for predicting multi-factor input models that are not time series.