Constructing a highly accurate price prediction model in real estate investment using LightGBM

Abstract

In this research, we propose a high-accuracy price prediction model for the purpose of constructing a support system for information collection and automatic analysis of profitable properties in the real estate investment market. In the traditional real estate investment process, investors needed to go through the following processes: 1) collect information on the Internet, 2) make price predictions based on their own judgement, 3) order, 4) negotiate and purchase. 1 and 2 in particular are inefficient because they seem simple, but are very time-consuming and must be repeated many times until a suitable property is found. Therefore, we aim to construct an efficient real estate investment support system by automating the information gathering process and substituting the price prediction process with a machine learning model. In this paper, we focus on the price prediction of part (2) and propose a highly accurate price prediction model using LightGBM. Specifically, the accuracy was improved by incorporating the condominium brand name, which is a price determining factor unique to Japan, and the Geo Data, a geographic factor, into the price prediction model.