A Rule-Based Approach to Developing Software Development Prediction Models

Managers of software development projects increasingly recognize the importance of planning and estimation and now have many sophisticated tools at their disposal. Despite this many systems are still delivered way behind schedule, cost far more to produce than original budget estimates and fail to meet user requirements. It is the contention of the authors that many existing tools are inadequate because they fail to embrace the significant body of knowledge accumulated by past and present project managers. This paper presents a new approach to planning which enables project managers to learn from the experience of others. The authors have adopted a bottom-up approach to planning which goes from the specific (planning the requirements capture and analysis process - RCA) to the general (planning the whole development process). A model, called MARCS, was constructed to give predictions of the resources (time, effort, cost, people) needed for the completion of and outcomes of the RCA process. Based on the predictions about the RCA process, the model then attempts to predict the resources and outcomes of the whole development process. MARCS is a combination of rule-based models and its main advantage is that it incorporates both qualitative and quantitative factors that can be easily identified and measured in the beginning of the development process. Empirical data concerning 107 projects developed by more than 70 organizations within UK, gathered through a two-stage mail survey was used for the construction and validation of the MARCS planning model.