Decision Making under Uncertainty in Software Engineering: Release Readiness

本研究探讨了软件缺陷的检测。这项研究的目的是解决“何时停止测试软件并发布它？”这个问题。在“立即交付！”由于时间和预算的限制，软件经理在现实生活中不会只观察一个因素，比如缺陷率。管理者将他们的先验知识与代表不同生命周期过程的因素及其对软件产品最终可靠性的影响的历史数据结合起来。此外，产品可靠性与软件开发生命周期（SDLC）活动密切相关。SDLC中的每个阶段都必须通过考虑产品、过程和人员相关因素之间的关系，以及这些阶段之间的关系以及它们对最终软件产品可靠性的影响来建模。先前关于发布计划的研究很好地解决了SDLC过程如何基于涉众偏好来确定优先级。然而，还需要对SDLC过程之间的偶然关系进行建模，以确定发布是否准备好了其可靠性。本研究旨在了解SDLC活动导致停止测试活动的因果关系。这个研究计划是基于实证分析的，目的是开发一个模型来量化软件产品的发布准备情况。软件工程领域的其他研究人员也可以使用该模型。我的目标是开发一个贝叶斯模型，该模型使用基于数据可用性和质量的定量和定性数据。这种混合模型将从组织系统中提取的本地数据与通过问卷收集的专家判断相结合，从而提供灵活性。我还计划扩展我现有的软件度量收集工具Prest，以包含更多的度量，以及用于分析、缺陷和可靠性预测的功能。拟议的研究计划与加拿大软件开发行业高度相关。加拿大的软件开发公司可以使用该工具来帮助他们的经理在不确定的情况下做出决策。

This research investigates the detection of software defects. The aim of the research is to address the question of "when to stop testing the software and release it?'. Under the constant pressure of "Deliver Now!' due time and budget constraints, software managers in real life do not observe only one factor, such as defect rates. Managers combine their prior knowledge with historical data about factors representing different lifecycle processes and their effects on the final reliability of software product. Furthermore, product reliability is strongly connected with software development life cycle (SDLC) activities. Each phase in the SDLC has to be modeled by considering the relations of product, process, and people related factors, as well as the relationships between these phases and their impacts on the final software product reliability. Previous studies in release planning well addressed how SDLC processes can be prioritized based on stakeholder preferences. However, there is also a need to model casual relations between SDLC processes to decide if the release is ready in terms of its reliability. This research aims to understand cause - effect relationships of SDLC activities leading decisions to stop testing activity. This research program is based on empirical analyses and aims to develop a model to quantify release readiness of software product. Other researchers in the software engineering domain can also use this model. My objective is to develop a Bayesian model that uses both quantitative and qualitative data based on data availability and quality. Such a hybrid model would give flexibility by combining local data extracted from organizational systems with expert judgments collected through questionnaires. I also plan to extend my existing software metrics collection tool, Prest, to include more metrics, and capabilities for analysis, defect and reliability prediction. The proposed research program is highly relevant to the Canadian software development industry. Software development companies in Canada to assist their managers to make decisions under uncertainty can use the tool.