| 摘要: | 營建產業中，契約協商之結果可能決定專案成本效率，甚至提升營造廠之執行績效。本研究採用分類與回歸決策樹 (Classification and Regression Tree, CART) 預測協商結果之議價次數與減價幅度，以期在採購階段促成契約議價協商。本研究首先自過去文獻彙整交易特徵影響因素與總結適當之預測模型，並且蒐集近五年共計 10,470 筆營建專案契約協商原始資料。資料顯示工程專案地點大多位於臺灣北部地區，其中 90%為總價合同類型;交易特徵方面，議價次數介於 1 次至 5 次之間，減價幅度分布於 0%至 94%;依工程特性分類，可分為假設工程、地工工程、結構工程、外牆工程、內裝工程、門窗工程、設備機電工程、景觀工程與雜項工程。在議價次數與減價幅度之預測結果方面，準確度範圍分別介於 79.9%至 95.5%以及介於 55.6%至 86.9%之間。由於CART 模型之預測結果分布範圍較廣，依據進一步探討之預測結果，本研究可發展出 4個構面之策略，分別為議價次數多且減價幅度高、議價次數多卻減價幅度低、議價次
數少卻減價幅度高、以及議價次數少且減價幅度低，以輔助營造廠於採購階段調整議價協商策略。
;Contract negotiation may determine cost efficiency and even enhance performance to the general contractors in the construction industry. The research objective is to facilitate contract negotiation in the procurement stage using classification and regression tree (CART) in order to predict the negotiation quantity and price deduction margin. Firstly, the literature review summarizes the impact factors and proper prediction model, followed by the data collection within recent 5 years, resulting in 10,470 raw datasets of contract negotiation for construction projects. These data present the project locations mainly in the north Taiwan and 90% among them with the lump sum contract type. The data range for negotiation quantity is from 1 to 5 times and for price deduction margin is from 0% to 94%. There are 9 project categories involving falsework, earthwork, structure, wall, interior work, window, facility installation, landscape, and miscellaneous work. The prediction results on both negotiation quantity and price deduction margin are with the accuracies from 79.9% to 95.5% and from 55.6% to 86.9%, respectively. Since the results yielded form CART vary in wide ranges, the findings are interpreted further to develop negotiation strategies for general contractors in 4 aspects of high negotiation quantity with high price deduction margin, high negotiation quantity with low price deduction margin, low negotiation quantity with high price deduction margin, and low negotiation quantity with low price deduction margin. |
