Size-Based Software Cost Modelling with Artificial Neural Networks and Genetic Algorithms

Abstract

Accurate software cost estimation has always been a major concern especially for people involved in project management, resource control and schedule planning. A high-quality and reliable development effort estimate could provide more efficient planning and control over the whole software process and guide a project to success. As literature shows many researchers proposed a plethora of methods and techniques to model the relationship between software size and the actual development costs (Jorgensen & Shepperd, 2007). However, the track record of IT projects shows that often a large number fails. Most IT experts agree that such failures occur more regularly than they should (Charette, 2005). According to the 10th edition of the annual CHAOS report from the Standish Group that studied over 40,000 projects in 10 years, it seems that success rates increased to 34% and failures declined to 15% of the projects. Even though success rates increased, still, 51% of the projects overrun time/budget, lack critical features and requirements and/or important quality requirements are compromised. Furthermore, average software costs are apparently overrun by 43% (Software Magazine, 2004). One of the main reasons for these figures is failure to estimate the actual effort required to develop a software project. A reliable software cost estimation model has always been a major challenge and demand for project managers at the initiation phase of the project and also an important asset for the whole process of software development. In addition, there is a large discussion on the discovery of the relationship between cost drivers and effort, especially of one of the most critical cost factors, namely software size (Sommerville, 2007). The aforementioned modelling and estimation problem is further amplified due to the high level of complexity and uniqueness of each project. Estimating software costs, as well as deciding on assessing the appropriate cost drivers, remain difficult issues that need to be addressed. Such issues are constantly at the forefront of the project management’s interests from the initiation of the project until the system is delivered. Cost estimates, even for well-planned projects, are hard to make and will probably concern project managers long before the problem is adequately solved. Over the years software cost estimation has attracted considerable research attention and many techniques have been developed to effectively predict software costs (Briand &