Classification of Software Projects using k- Means, Discriminant Analysis and Artificial Neural Network

An attempt is made in this paper to identify the groups of software development projects which demonstrate the significance of comparable characteristics based on various parameters associated with the Source Lines of Code (SLOC). Initially, software projects are clustered by using k-means cluster analysis. These groups are investigated further with the Discriminant Analysis (DA) and compared with Artificial Neural Network (ANN). For this purpose, we collected the historical data of software development projects from one of the major information technology (IT) company and this provided essential progressive information such as Planned Value (BCWS), Actual Cost (ACWP), Earned Value (BCWP), Cost Performance Index (CPI), Average Team Size (Team_Size) and Source Lines of Code (LOC).The results of this comparison study indicate that a Statistical model based on discriminant analysis is marginally better for prediction of the effort, than a non-parametric model based on artificial neural network. The classification method proposed in this paper may be used to identify the similar category of projects and for forecasting the software development cost and time effort. Hence, this approach would be useful for planning and preventive actions in the process of software development. The Statistical Software Package IBM SPSS 19.0 is used for the present analysis. Index Terms — Artificial Neural Network, Cluster Analysis, Discriminant Analysis, k-Means, Lines of Code, software development estimate and Software Effort Estimation. ——————————  ——————————