本研究以臺南市東區土地交易市場為研究對象，探討在營建環境變動下，土地價格的影響機制與預測方法。透過決策樹模型，建立具系統性的土地價格預測方法，提升投資決策的準確性。模型演算採用CART與CHAID演算法，分別以基尼指數與卡方檢定作為分裂準則，提升模型預測能力與解釋效果。土地交易資料以內政部實價登錄資料為基礎，整合公告地價、建蔽率、容積率、臨路條件及經濟指標等變數，相關變數皆來源於政府公告網站。透過SAS Enterprise Miner進行資料處理與模型建構。資料經篩選後，以70%作為訓練集、30%作為測試集，並透過R-squared、MAE、RMSE與MAPE等指標評估模型表現，以綜合判斷模型之穩定性與預測能力。
研究結果顯示，不同決策樹模型在預測能力上存在明顯差異。CHAID五元決策樹的解釋力最高(R2約0.814)，對資料分群能力較佳，但其MAE為7.470萬元/每坪、RMSE為9.184萬元/每坪，MAPE為21.79%，這些指標顯示預測誤差較大。在預測誤差方面，CHAID三元決策樹表現最佳，其MAE為5.435萬元/每坪、RMSE為6.447萬元/每坪，MAPE為17.42%，預測值與實際成交價格的偏差最小、最準確，同時R²達0.764，亦具良好的解釋能力，顯示在「解釋能力與預測精度」之間取得較佳平衡。綜合評估各項指標，CHAID三元決策樹為最適合之土地價格預測模型。以實際臺南市東區等兩塊售賣中土地作為預測舉例對象，進行土地交易價格預測，預測成果顯示3年內土地交易價格幾乎沒有變動，代表土地的增長幅度於短時間內不明顯。
決策樹模型節點分裂結果顯示，公告地價為最關鍵的影響變數，其次為營建物價指數、消費者物價指數、臨路面寬、建蔽率與等土地開發條件。決策樹模型能有效辨識不同價格區間之分群特性，具備良好解釋性與實務應用價值，可作為開發商與投資人評估土地價值與制定投資策略之重要參考依據。舉例應用，進行土地切割時，盡可能保留臨路面寬大於22m。收購整合零碎土地時，加權之公告地價保持3.352、7.235萬元/坪，在土地價格預測上可取得單位的最大價值。

This study focuses on the land transaction market in the East District of Tainan City, aiming to explore the mechanisms influencing land prices under a changing construction environment and to construct an accurate predictive model. In recent years, driven by the development of the Southern Taiwan Science Park and the influx of high-tech industries, land prices in this area have risen significantly. However, issues such as soaring post-pandemic construction costs and severe labor and material shortages have substantially increased land development risks. Given that traditional, experience-based valuation methods struggle to keep pace with a rapidly changing market, this study introduces decision tree models to establish a systematic and objective approach for land price prediction, thereby enhancing the accuracy of investment decision-making.
In terms of methodology, this study employs both CART and CHAID algorithms, utilizing the Gini index and chi-square test as their respective splitting criteria. The primary data source is the Actual Price Registration system of the Ministry of the Interior, integrated with variables including announced land value, building coverage ratio, floor area ratio, road facing conditions, and macroeconomic indicators. Modeling was conducted using SAS Enterprise Miner. The dataset was divided into training and testing sets at a 7:3 ratio. Model performance was comprehensively evaluated using metrics such as R-squared, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE).
Empirical results indicate that the CHAID five-branch decision tree exhibits the highest explanatory power (R² ≈ 0.814). However, when comprehensively considering prediction error and model stability, the CHAID three-branch decision tree performs the best. It achieved an MAE of 5.435 (10,000 NTD/ping), an RMSE of 6.447 (10,000 NTD/ping), and a MAPE of 17.42%, alongside a strong R² of 0.764. By striking the optimal balance between explanatory power and predictive accuracy, it is identified as the most suitable model for this study.
Further node splitting analysis reveals that "announced land value" is the most critical variable influencing land prices, followed by development attributes such as road facing conditions and building coverage ratio. This indicates that land prices reflect not only locational conditions but also rely heavily on land development potential. In conclusion, the decision tree model effectively identifies price clustering characteristics and possesses high interpretability. It can serve as a crucial practical reference for developers and investors in conducting land valuation and risk management.

1.Shannon, C. E., A Mathematical Theory of Communication, Bell System Technical Journal, 27, 3, 379–423, 1948.
2.Kass, G. V., An Exploratory Technique for Investigating Large Quantities of Categorical Data, Applied Statistics, 13, 2, 119–127, 1964.
3.Hunt, E. B., Marin, J., & Stone, P., Experiments in Induction, Academic Press, New York, 1966.
4.Rosen, S., Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition, Journal of Political Economy, 82, 1, 34–55, 1974.
5.Quinlan, J. R., Discovering Rules by Induction, Machine Learning: An Artificial Intelligence Approach, McGraw-Hill, New York, 1979.
6.Breiman, L., Classification and Regression Trees, Wadsworth, Belmont, CA, 1986.
7.Brueckner, J. K., The Structure of Urban Equilibria: A Unified Treatment of the Muth-Mills Model, Handbook of Regional and Urban Economics, 2, 821–845, 1987.
8.Quinlan, J. R., C4.5: Programs for Machine Learning, Morgan Kaufmann, San Mateo, CA, 1993.
9.Cheshire, P., & Sheppard, S., On the Price of Land and the Value of Amenities, Economica, 62, 246, 247–267, 1995.
10.Colwell, P. F., & Munneke, H. J., The Structure of Urban Land Prices, Journal of Urban Economics, 41, 3, 321–336, 1997.
11.Colwell, P. F., & Munneke, H. J., Land Prices and Land Assembly in the CBD, Journal of Real Estate Finance and Economics, 18, 2, 163–180, 1999.
12.Friedman, J. H., Greedy Function Approximation: A Gradient Boosting Machine. The Annals of Statistics, Vol. 29, No. 5, Institute of Mathematical Statistics, Beachwood, OH, USA, pp. 1189–1232, 2001.
13.Des Rosiers, F., Power Lines, Visual Encumbrance and House Values: A Microspatial Approach to Impact Measurement, Journal of Real Estate Research, 23, 3, 275–301, 2002.
14.Quigley, J. M., & Rosenthal, L. A., The Effects of Land Use Regulation on the Price of Housing: What Do We Know? What Can We Learn?, Cityscape, 8, 1, 69–137, 2005.
15.Sirmans, G. S., Macpherson, D. A., & Zietz, E. N., The Composition of Hedonic Pricing Models, Journal of Real Estate Literature, 13, 1, 3–43, 2005.
16.Han, Kamber & Pei, Data Mining: Concepts and Techniques (3rd ed.). Morgan Kaufmann, Waltham, MA, USA, 327–360, 2012
17.Hair, J. F., Hult, G. T. M., Ringle, C. M. & Sarstedt, M., A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) (3rd ed.). SAGE Publications, Thousand Oaks, CA, USA, pp. 209–219, 2022.
18.簡淑芬，不動產估價模型之研究－以臺中市為例，碩士論文，逢甲大學，2012。
19.馮世傑，房價影響變數之探討－以台北市為例，碩士論文，東吳大學，2014。
20.趙健宏，以區域經濟與環境特徵建構不動產估價趨勢函數，碩士論文，國立中央大學，2015。
21.葉紫光，影響房價因素之研究－不動產估價技術規則的觀點，碩士論文，天主教輔仁大學，2018。
22.張琪蓉，不動產估價模型建立－以臺北市東西區公寓為例，碩士論文，國立臺灣科技大學，2018。
23.蕭弘偉，臺北捷運文湖線人口計土地價格波動特性分析，碩士論文，國立臺北科技大學，2019。
24.李政育，應用隨機森林預測高雄市房價，碩士論文，國立東華大學，2019。
25.蔡尚恩，應用集群分析及隨機森林建構房地產價格模型-以高雄市為例，碩士論文，國立交通大學，2020。
26.曾炳耀，地價與建物型態關聯性之研究－以竹北市都市計畫區為例，碩士論文，明新科技大學，2020。
27.陳冠廷，以類神經網路預測台南市東區的房價，台南市成大城房價之預測，碩士論文，國立成功大學，2020。
28.許文昌，土地經濟學體系，元照出版公司，101–114，2021。
29.黃智穎，台南市成大城房價之預測，碩士論文，國立成功大學，2021。
30.邱辰昀，應用卷積神經網路合併梯度提升決策樹模型於多重輸出問題之工業大數據預測分析，碩士論文，國立臺灣科技大學，2021。
31.柯昆緯，結合類神經演算法及地理資訊系統以建立不動產估價模型－以臺南市永康大灣透天住宅為例，碩士論文，崑山科技大學，2021。
32.沈芝妘，地價及標準地價評議委員會審議制度之探討，碩士論文，國立臺北大學，2021。
33.包晃豪，決策樹運用於工程查核選案之研究，碩士論文，國立臺灣科技大學，2021。
34.許博榮，建構金門地區房地產決策樹分類應用模式，碩士論文，國立金門大學，2021。
35.樊俊文，梯度提升決策樹回歸模型在房價預測中的性能比較與優化，碩士論文，長庚大學，2022。
36.陳伯杰，比較隨機森林和XGBoost的預測強韌性，碩士論文，淡江大學，2022。
37.王尹暘，以決策樹預測台南世紀之門房價，碩士論文，國立成功大學，2023。
38.許庭瑄，比較高鐵站不同選址地點對周圍住宅價格之影響：特徵價格模型的多層次分析，碩士論文，國立臺灣大學，2023。
39.李佩臻，應用集成學習建立房價預測模型，碩士論文，中原大學，2023。
40.李祖源，運用決策樹和類神經網路預測台灣鋼筋價格，碩士論文，國立成功大學，2024。
41.柯博仁，台灣高速公路鋪面養護工法決策樹建立之研究，碩士論文，國立成功大學，2024。
42.陳爰棻，台積電宣布投資高雄楠梓對楠梓區地價之影響分析，碩士論文，國立高雄科技大學，2024。
43.施柏佑，桃園市重劃區與舊城區房價影響因素之研究，碩士論文，國立中央大學，2024。
44.詹仁廷，臺北市宗教設施對周遭房價之影響－以特徵價格法與雙重差分法評估，碩士論文，國立臺北大學，2024。
45.胡祺嚴，以隨機森林法預測臺南市桂田磐古社區之房價，碩士論文，國立成功大學，2025。
46.陳忠彥，以決策樹與迴歸分析法預測嘉義科學園區周邊的房價，碩士論文，國立成功大學，2025。