Designing Software Development Processes to Optimize Multiple Output Performance Characteristics

摘要

Quantitative project management has become essential for the successful execution of software development projects. Successful execution implies that quality software is delivered on time, without cost overruns (which includes overtime). Project managers need to execute the projects in such a way that the goals for quality, productivity, level of effort, cost, etc., are all met. Many of the aforementioned characteristics can be correlated with one another and thus can be difficult to manage independently. The literature on quantitative project management generally deals with monitoring a single performance characteristic, most often software quality, as the response variable. In this article, the authors demonstrate how the quantitative project management methodology can be used to identify a common process setting or team composition to simultaneously achieve the goals for multiple characteristics. This is achieved by integrating multiple response optimization methodology with quantitative project management. Using the case-study approach, multiple response optimization for quantitative project management is performed for an agile software development scenario in the information technology industry in India. The model is used to predict two output characteristics selected by project managers: defect density and sprint velocity. The model was created using data from 90 sprints, with classification and regression trees (CART) using the mean square error criterion as the optimization approach. The result from multiple response optimization is shown to be superior to that of controlling the individual performance characteristics separately, and the results are validated on six projects. Even though only two characteristics are considered in the case study, the methodology can be applied to simultaneously monitor many more characteristics.