建構軟體測試階段之累積失效時間預測模型

Abstract

相較於應用複雜的系統軟體，手機的應用程式軟體開發屬於較小型的軟體專案，在此類軟體開發之專案管理方面，需同時考量品質與時程兩個指標，然軟體開發人員為了取得市場的先機，在產品測試階段，通常優先考量的指標是時程，在此情況下，如何在品質與時程兩者中取得帄衡，是一個重要的議題。為確保軟體之品質，軟體可靠度是一個重要的衡量軟體品質之指標，正確的預測軟體可靠度可以確保軟體的品質，且軟體可靠度的預測值可以幫助軟體開發人員評估軟體品質是否達到預期目標、所需測試時間以及預估何時可結束測試階段，進而掌握軟體發行時程。軟體可靠度模型大致上可分為兩類：解析模型(analytical model)和資料導向模型(data-driven model)。近年來，資料導向模型在軟體可靠度模型中越來越受重視，其內容是針對軟體累積失效時間做預測，由於資料導向模型不需要對軟體失效過程做任何統計假設，只要根據所觀測到的軟體失效的數據，將其視為一個時間序列即可進行建模與預測。本研究針對軟體發展生命週期中的測試階段，發展一個軟體累積失效時間預測模型，用來預測軟體下一次發生失效的時間點。本研究所建之預測模型不需要大量建模資料，也不需符合任何統計假設，建構過程相當簡便，軟體開發人員和測試人員在測試過程中的初期階段，就可獲得軟體累積失效時間相關資訊給管理者做決策，以適時將產品推入市場。最後本研究利用三個軟體失效案例，來說明本研究方法之準確性。

Software development for mobile phone applications, considered as a small project involves less complicated system than that of other application software. In such software development, the project management considers two indicators: quality and schedule. However, developers usually emphasize on schedule in order to obtain market opportunities in product testing phase. In this case, an important issue is that how to strike a balance in both the quality and schedule. To ensure software quality, software reliability is an important measure of software quality indicators. The prediction of software reliability is important to the project management in order to support decision making such as release time for the product releases. Software reliability models can be broadly divided into two categories, analytical model and data-driven model. In recent years, data-driven model of software reliability models received more attentions in recent years. The data driven model does not require any statistical assumptions on software failure process. Thus, they may be carried out like a time series model. The main object of this study is to apply rolling grey model to predict the next cumulative failure time occurs in the software testing phase. This model needs a small number of data and no statistical assumptions. The result of the prediction model can help managers to timely decide the right timing of software release based on cumulative failure time. Finally, this study uses three software failure cases to illustrate the accuracy of the method.