Monte Carlo Simulation for Software Maintenance Cost Estimation

All software estimation techniques rely in some way on past experience to predict the time and effort required to accomplish a software task. The accuracy of these techniques depends on 1. the extent to which past experience is relevant to the a new project at hand, 2. the extent to which any variation has been accounted for and parameterized, and 3. the skill of the user estimator in assigning values to parameters. In this paper, we explore the use of in-house organization's metrics for the purpose of obtaining an estimation tool method that is both more accurate and easier to implement. This approach is evaluated using existing data from software maintenance programs of embedded systems at General Dynamics. We develop a simple, flexible and accurate model using historical data and simulation modeling. The model is simple because it is mostly based on extrapolation from historical data, additional parameters are added in a graphical way. It is flexible because the simulation modeling tool allows users to account for a variety of parameters. It is accurate because the simulation modeling tool allows users to account for as many parameters as needed.