臺灣訂製型軟體成本估算模式之研究

Abstract

軟體成本估算提供專案管理的基礎。精確的成本估算需要經常性的經驗累積，將累積的經驗組織成更精煉的成本估算模式，以提高估算的精確度。而軟體成本估算模式多以軟體規模輸入估算方程式，以估算出發成本，其準確性與預估者的經驗、預估者對專案的瞭解程度、及成本模式估算的適用性有關。 本論文透過與十位軟體業中高階管理者的訪談方式，瞭解臺灣軟體業成本估算的現況，顯示目前仍停留在以個人經驗法則作為估算依據，缺乏適合的軟體成本估算模式及相關資料的收集。本論文選擇成本估算合成模式-COCOMO模式，作為資料收集的參考依據，經由問卷調查方式，收集臺灣軟體業的訂製型軟體發展成本資料，找出影響臺灣軟體發展成本的重因素及重要程度排名，以平均相對誤差（MRE-）、預測水準（PRED）、及相關係數（CORR）三種衡量尺度測試COCOMO模式的適用性，發現多高估軟體成本，適用性不高。可能的變異來源，有些源自於COCOMO模式本身的限制，有些則是因為臺灣軟體發展環境的限制。此外，以迴歸方法修正COCOMO模式之參數，發現臺灣的軟體成本估算方程式指數小於1，呈現規模經濟性。經由對軟體成本估算模式的修正，提供適合國內軟體開發環境的成本估算模式。

Software cost estimation provides the basis of the project management. Better estimation often needs historical project data and managers' experince to refine a cost estimation model. It inputs the project size formula to estimate the development effort. The estimater's experience, understanding to the project,and the portability of the cost estimation model also influence the accuracy. At first, we interview with ten software top managers to investigete the current situation on software cost estimation in Taiwan, The results shows that they do not use any software cost estimation model. The estimation generally depends on prersonal experience. The project development and relevant data are also insufficient to build a cost estimation model. Additionally, this study collects 35 projects' actual data to test COCOMO model's portability in Taiwan and find proper cost drivers. Three metrics, MRE-(mean relative error)、PRED (prediction leved) and CORR (correlation), are used to test COCOMO model's portability. The testing results shows portability is very low and often overestimated development cost. Some variations may come from the COCOMO model's restrication or the uniqueness of Taiwan's development environment. Besides, by using regression method to calibrate the coefficient less than 1, this study augmented the COCOMO model to provide a adaptive cost estimation model.