Models for Predicting Development Effort of Small-Scale Visualization Projects

Abstract

Software project effort estimation is one of the important aspects of software engineering. Researchers in this area are still striving hard to come out with the best predictive model that has befallen as a greatest challenge. In this work, the effort estimation for small-scale visualization projects all rendered on engineering, general science, and other allied areas developed by 60 postgraduate students in a supervised academic setting is modeled by three approaches, namely, linear regression, quadratic regression, and neural network. Seven unique parameters, namely, number of lines of code (LOC), new and change code (N&C), reuse code (R), cumulative grade point average (CGPA), cyclomatic complexity (CC), algorithmic complexity (AC), and function points (FP), which are considered to be influential in software development effort, are elicited along with actual effort. The three models are compared with respect to their prediction accuracy via the magnitude of error relative to the estimate (MER) for each project and also its mean MER (MMER) in all the projects in both the verification and validation phases. Evaluations of the models have shown MMER of 0.002, 0.006, and 0.009 during verification and 0.006, 0.002, and 0.002 during validation for the multiple linear regression, nonlinear regression, and neural network models, respectively. Thus, the marginal differences in the error estimates have indicated that the three models can be alternatively used for effort computation specific to visualization projects. Results have also suggested that parameters such as LOC, N&C, R, CC, and AC have a direct influence on effort prediction, whereas CGPA has an inverse relationship. FP seems to be neutral as far as visualization projects are concerned.

Keywords

• multivariate liner regression
• neural networks
• nonlinear regression
• predictive models
• software development effort
• visualization projects