Evaluating software development effort model-building techniques for application in a real-time telecommunications environment

摘要

This paper describes the comparative evaluation of four methods of building software development effort models based on least squares regression, artificial neural networks, case-based reasoning and rule induction. Some deficiencies are identified in the main measurement of estimating effectiveness currently used in comparative evaluations, 'mean magnitude of relative error' (MMRE), and a complementary measurement, 'mean variation from estimate' (MVFE) is suggested as more accurately reflecting the practitioner's viewpoint. Given the current state of development of the techniques, the parallel use of least squares regression and case-based reasoning is recommended as appearing to give the most reliable results in the studied environment.