PCA-assisted Regression Models for Software Development Effort Prediction: Small-scale Visualisation Projects

Effort estimation is a process of predicting probable cost and development time of a software. Estimating software development effort remains a complex problem, and the one which continues to draw significant research attention. Correctness in estimating the required software development effort plays a critical factor in the success of software project management. For good software estimation model, the estimated effort should be approximately equal to the actual effort. Accurate estimation allows manager to allocate the resources to plan and coordinate all activities. Several techniques like neural network, fuzzy logic, genetic engineering and regression techniques are used either individually or in combination as hybrid approaches to predict the effort. In this work, the effort estimation for small-scale visualisation projects developed by postgraduate students in supervised academic setting is done in following stages: (i) Elicitation of seven software parameters that are novel in nature namely lines of code (LOC), new and changed code (N&C), cumulative grade point average (CGPA), reuse code (R), cyclomatic complexity (CC), algorithmic complexity (AC) and functional points (FP), (ii) dimensionality reduction in the number of parameters using Principal Component Analysis (PCA) and, finally, (iii) development of linear regression and polynomial regression models using the parameters derived from PCA. Evaluations of the regression models have shown mean error to be 6.275536e-16 and 5.72875e-16, respectively, for multiple linear regression and polynomial regression models during verification. Thus, the marginal differences in the error estimates have indicated that the two models can be alternatively used for effort prediction specific to small-scale visualisation projects considering only three significant attributes namely LOC, CGPA and N&C.