A Long Short-Term Memory Model using Kernel Density Estimation for Forecasting Apartment Prices in Seoul City

Predicting housing prices is a significant issue for various stakeholders, ranging from individuals, organizations, and governments. Previous studies have focused on predicting apartment prices using statistical methods such as regression analysis, but recently, research using machine learning and machine learning has been actively conducted. Also, some studies have applied convolutional neural networks using image information and time series analysis methods for prediction of housing prices. Factors affecting housing prices are diverse and are affected by regional characteristics and location. In particular, since housing prices are time series characteristics and affected by trends, a prediction model that takes this into account is necessary. Thus, this paper utilized density information of commercial and convenience facilities around houses using long short-term memory (LSTM) model with kernel density estimation information to predict apartment prices in Seoul. The training data was from January 2020 to December 2023, and the test data was from January 2024 to September 2024. We obtained and preprocessed real apartment transaction data, regional gross domestic product index, population density, and demographic data by age group from various websites. In particular, this paper used extracted density information through kernel density estimation method after visualizing the number of commercial facilities and various convenience facilities around the apartments and used the unit price per area of individual apartments for a target variable. Then, this paper compared the prediction results of various methods such as regression analysis, deep neural network (DNN), and long short-term memory (LSTM) models. The results of this paper showed that the prediction performance was high when the number or density of convenience facilities was reflected in each model, and the prediction performance was the best when the density of convenience facilities through kernel density estimation was utilized. In particular, LSTM which reflects the time series nature, showed higher predictive performance than any other models such as regression and DNN. This paper suggests that density information that visualizes convenience facilities around houses is useful and time series analysis methods such as LSTM that can reflect trends give more suitable for forecasting housing prices such as apartment.