Title Investigating Problem Analysis and Solution Strategies in Predicting Housing Price IndexThrough Sequential Application of Transformer and LSTM Models
Authors 윤도경(Yoon, Do-Kyung) ; 신동윤(Shin, Dong-Yoon)
DOI https://doi.org/10.5659/JAIK.2024.40.1.35
Page pp.35-44
ISSN 2733-6247
Keywords LSTM; Transformer; Housing Price Prediction; Problem Analysis; Solution Strategies; Machine Learning; Predictive Modeling; Error Analysis
Abstract This study aims to review previous research on factors affecting the housing price index and construct a prediction model for the index using Long Short-Term Memory (LSTM) and Transformer models. Specifically, it combines LSTM, specialized in processing time-series data, and DistilBERT, specialized in handling text data, to utilize both historical housing price index data and relevant news articles. The experimental results of the proposed model confirmed significant accuracy when comparing predicted values in each region (J, S, G) with the actual values. However, some clusters displayed relatively high errors, indicating a need for additional analysis and improvement. Additionally, it was observed that subjective elements could significantly impact the interpretation of clustering results, highlighting the necessity for further analysis. Result visualization and statistical analysis were conducted, confirming their accurate reflection of housing price fluctuation trends in each region. This study introduces a novel approach to predicting the housing price index using deep learning models like LSTM and DistilBERT, providing valuable insights into real estate market trend predictions. The approaches and findings from this research are anticipated to provide valuable starting points for further exploration of creative solutions and the development of effective problem-solving strategies.