預測台灣房地產市場趨勢之模型---應用深度學習技術Prediction Model for the Trend of Taiwan’s Real Estate Market—Applying Deep Learning Technology

本文以人工智慧之深度學習技術,預測台灣房地產之趨勢,除了採用總體經濟變數外,亦運用網路搜尋量與人口結構變數以進行房地產市場之預測,透過LSTM (Long Short-Term Memory)預測房地產市場之趨勢,並與GARCH (Generalized Autoregressive ConditionalHeteroskedasticity)模型進行預測能力準確度之衡量。結果顯示LSTM模型對台灣房地產市場之預測結果優於GARCH模型,證實LSTM預測模型具有預測房地產市場趨勢之可行性。最後,採用滾動視窗逐年調整LSTM模型參數,在台灣房價指數與成交量指數預測之準確度分別提升至97%與76%,更能有效提高其模型預測能力,經過滾動視窗之調整後,亦具備與時俱進之效果,預測結果能更貼近現實狀況。
關鍵詞:房地產市場、深度學習、網路搜尋量、人口結構、長短期記憶模型

ABSTRACT

This study used artificial intelligence deep learning technology to forecast the real estate trends in Taiwan. Economic variables, internet search volume, and demographic variables are utilized to forecast the real estate market. Furthermore, the real estate market trends were predicted based on the Long Short-Term Memory (LSTM) model and the accuracy of the predictions was evaluated using the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model. The results reveal that the LSTM model predicted the Taiwan real estate market better than the GARCH model, thus verifying the feasibility of the LSTM model in forecasting the trend of real estate markets. Furthermore, a rolling window was used to adjust the parameters of the LSTM model annually. Therefore, the prediction accuracy of Taiwan’s house price and transaction volume indices improved to 97% and 76%, respectively, thereby enhancing its model prediction ability. Moreover, adjusting the rolling window also enabled the advancements over time; therefore, the predicted results were closer to the practical situation.

Key words: Real Estate Market, Deep Learning, Searching Volume, Demographic Variables, Long Short-Term Memory Model