房屋價格分析：以高雄市左營區為例

Abstract

中文摘要 近年，隨著高雄市發展重心北移、台積電設廠及三鐵共構之交通優勢，左營區已成為北高雄商業與交通核心，房地產市場結構隨之發生改變。本研究旨在探討民國109年11月至114年11月間，高雄市左營區住宅大樓、華廈及公寓之房價波動與影響因素，研究透過Python網路爬蟲技術，自「內政部不動產交易實價查詢服務網」擷取共10,993筆有效交易數據，進行資料清洗、特徵工程與離群值處理，以確保數據品質。 本研究結合敘述性統計、地理空間分析與機器學習演算法，比較線性迴歸、決策樹、隨機森林及梯度提升樹四種模型對房價的預測能力。研究結果顯示，隨機森林模型表現最佳，其決定係數(R^2)達0.8409，均方根誤差（RMSE）為35,778元/坪，顯著優於傳統線性迴歸模型之決定係數結果0.53，說明房價影響因素間存在高度非線性關係。 特徵重要性分析指出，「屋齡」為影響左營區房價之關鍵因素，重要性佔35.85%，其次為「交易年份」之25.11%與「地理位置」，經緯度合計為16.59%。空間分析則說明以高鐵左營站為核心之同心圓價格梯度現象：核心區，如高鐵特區、蓮池潭首排之單價最高，平均落在35萬元/坪以上；巨蛋商圈及核心市區次之；外圍與老舊社區則價格相對較低。本研究透過數據，量化各項房屋屬性與外部環境對價格的影響，驗證新屋溢價與地段效應，並建立房價預測模型。

Abstract In recent years, driven by the northward shift of Kaohsiung City's development focus, the establishment of TSMC's manufacturing facilities, and the transportation advantages of the three-railway terminal, Zuoying District has emerged as the commercial and transportation hub of Northern Kaohsiung. Consequently, the structure of its real estate market has undergone significant transformation. This study aims to investigate housing price fluctuations and determinants for residential high-rises, mid-rise elevator buildings, and walk-up apartments in Zuoying District from November 2020 to November 2025. Utilizing Python web crawling techniques, 10,993 valid transaction records were extracted from the Ministry of the Interior's Real Estate Actual Transaction Price Query Service. Data cleaning, feature engineering, and outlier processing were performed to ensure data quality. Integrating descriptive statistics, geospatial analysis, and machine learning algorithms, this study compares the predictive capabilities of four models: Linear Regression, Decision Tree, Random Forest, and Gradient Boosting Trees. The results indicate that the Random Forest model achieved the best performance, yielding a coefficient of determination (R^2) of 0.8409 and a Root Mean Square Error (RMSE) of 35,778 TWD/ping. This significantly outperforms the traditional Linear Regression model (R^2 = 0.53), demonstrating the existence of highly non-linear relationships among housing price determinants. Feature importance analysis identifies "Building Age" as the most critical factor influencing housing prices in Zuoying District, accounting for 35.85% of importance, followed by "Transaction Year" (25.11%) and "Geographic Location" (Latitude and Longitude combined, totaling 16.59%). Spatial analysis reveals a concentric price gradient centered on the Zuoying High Speed Rail Station. The core zone, including the HSR special zone and the Lotus Pond waterfront, commands the highest unit prices, averaging over 350,000 TWD/ping. This is followed by the Kaohsiung Arena commercial district and the city center, while peripheral areas and older communities exhibit relatively lower prices. Through data analysis, this study quantifies the impact of various property attributes and external environmental factors on prices, verifying the new house premium and location effects, and establishes a robust housing price prediction model.