PREDICTING SOFTWARE QUALITY, DURING TESTING, USING NEURAL NETWORK MODELS: A COMPARATIVE STUDY

Abstract

The application of statistical modeling techniques has been an intensely pursued area of research in the field of software engineering. The goal has been to model software quality and use that information to better understand the software development process. Neural network modeling methods have recently been applied to this field. The results reported indicate that neural network models have better predictive quality than some statistical models when predicting reliability and the number of faults. In this paper, we will investigate the application of principal components analysis to neural network modeling as a way of improving the predictive quality of neural network quality models. Using data we collected from a large commercial software system, we developed a multiple regression model using the principal components. Then, we trained two neural nets, one with raw data, and one with principal components. Then, we compare the predictive quality of the three competing models for a variety of quality measures.