Software Product Effort Estimation with Sequence Diagrams and Regression Analysis

Software development effort prediction is a critical aspect of project management. In order to successfully complete a product within all the constraints, the forecasted effort must be fairly reliable. In this work, we have come up with a reasonably viable work-effort estimation for projects developed through Object Oriented Analysis and Design (OOAD) Approach. The approach proposed in this work uses a selected set of features present in the Unified Modeling Language (UML) Sequence diagrams and customized ‘Machine-Learning Regression-analysis (ML-RA)’ programs to come up with a viably accurate prediction. The ML techniques include: Extreme Gradient Boosting (EGB), Extreme Learning Machines (ELM), Decision Tree (DT), Bayesian Ridge Regression (BRR) and Gaussian Processes with Neural Network Kernels methods (GPNN). Based on the out-comes of our experiments, it was clearly evident that the GPNN approach provided the highest accuracy with respect to other ML models.