| Models of software processes, products, and resources are used to help predict, assess, understand and control project development. Choosing the appropriate set of models can be a difficult task, with a wide variety of factors influencing the decision, including the potential knowledge to be gained from the model, the accuracy and reliability of the model, and the ease of application and refinement of the model. The Experience Factory is an organizational structure aimed at improving software activities through the accumulation, analysis and packaging of a wide variety of software assets and experiences. We propose a refinement of the Experience Factory providing a capability supporting the creation, use, and maintenance of empirically based models. We suggest a generative, compositional approach to model construction based on evolving project characterizations. The approach offers the potential for more accurate predictions, improved assessment of uncertainty, and is amenable to automation. We use empirical analyses from the NASA/GSFC SEL in the areas of effort and quality modeling to provide evidence to the feasibility and benefit of such an approach. |
