應用梯度提升樹於實價登錄之車位價格預測

建立實價登錄制度之目的為解決不動產交易價格資訊不透明的問題，透過填報地址、總價、總面積、屋齡等多項交易資訊，可應用於各類不動產交易應用分析。在實價登錄的資料中，多數資料會將車位價格及房屋價格合併登記，但部份實價登錄資料會將車位價格與房屋價格分離，有助於理解車位價格單價。本研究應用基於決策樹之機器學習模型—梯度提升樹（Gradient Boosted Tree,GBT），利用有車位價格的實價登錄資料，使用實價登錄資料建立車位價格預測模型。於去除離群值後進行訓練，預測車位價格之均方根誤差為每個車位$236,653，以車位價格平均值$1,672,019而言，相對誤差約為14.15％。機器學習方法可將車位價格與房屋價格分離，有助理解實際房屋與車位價格。期望預測車位價格可在未來應用於房地產市場分析等多種用途。

The Real Price Registration system aims to address the issue of opacity in real estate transaction prices by reporting various transaction information such as address, total price, total area, and age of the property for various applications. In the data of the Real Price Registration system, most of the data combines the prices of parking spaces and houses, but some of the data separates the prices of parking spaces from those of houses, which helps understand the independent prices of parking spaces. In this study, we applied the Gradient Boosted Tree (GBT) machine learning model based on decision trees to predict the prices of parking spaces using Real Price Registration data with parking space prices. After removing outliers, the prediction accuracy for parking space prices can reach about NT$236,653 per parking space. Given the average parking space price of NT$1,672,019, the relative error is approximately 14.15%. This method can effectively separate the prices of parking spaces from those of houses, facilitating the understanding of parking space and the house prices. Expected that predicting parking space prices can be applied to various applications in urban planning and real estate market analysis in the future.