Soft Computing Approach To Some Software Metrics

Abstract

Software Planning begins before the technical work starts, continues as the software evolves from concept to reality and culminates only when the software is retired. The size is the key parameter for estimation of other activities. It is input to all costing models for the estimation of cost, development time and schedule for the project. If size is not estimated properly it may have a serious impact on other resource and estimation activities. The lines of codes for software are only known after the software is built. Therefore, to estimate the lines of code, once the functionality is known, is a challenging task. This thesis presents a few models to estimate size in terms of lines of code once the function point count is known. It also proposes a fuzzy approach to the function point analysis method. The fuzzy model can capture the inherent fuzziness present in all parameters. It explores the possibility of using a neural network model for estimating lines of code. Robust regression technique has also been effectively used to predict lines of code. The thesis further proposes an expert committee model, which is a combination of robust regression, and a neural network model. Bayesian regularization is used in the neural network model. This model is also evaluated in this thesis for various training algorithms to find out which one is the most suitable. The models proposed here are validated with empirical data of projects available from International Standards Benchmarking Software Group (ISBSG). newline... newline