ANALYSIS OF THE INTERNATIONAL SOFTWARE BENCHMARKING STANDARDS GROUP (ISBSG) REPOSITORY USING NEURAL NETS

摘要

Portfolio management and optimization techniques are becoming increasingly important in industry. As the number of project opportunities grows the corporate or financial planner is faced with an increasingly difficult task. Mathematical search and risk optimization algorithms can greatly simplify the planning process. A particularly well suited class of algorithms for optimizations are Neural Nets. This paper will discuss the application of Neural Nets in order to selecting the optimum set of project. We will show that Neural Nets are excellent at handling a large amount of data. This class of algorithms is capable of scaling upwards of a thousand projects, well beyond the reach of traditional methods. This will be shown with a small repository, yet realistic set of IT projects, which will be analyzed on a personal computer. Traditional approaches are founded on ranking models. Most models are based on limited number of factor. But identifying an appropriate set of factors can be a time consuming process. Neural Nets are capable of generating multiple sets, providing an opportunity to explore alternative characteristics of the portfolios. A portfolio analyst can request to belong projects to classes that satisfy the business objectives. We will show that factors identify a particular project. This can be help to match a project to existing portfolio project.