USING NEURAL NETWORKS TO PREDICT SOFTWARE FAULTS DURING TESTING

Summ. & Conclusions - This paper investigates the application of princ ipal components analysis to neural-network modeling, The goal is to pr edict the number of faults, 1. Ten software product measures were gath ered from a large commercial software system, Principal components wer e then extracted from these measures, 2. We trained two neural network s, one with the observed (raw) data, and one with principal components . 3. We compare the predictive quality of the two competing models usi ng data collected from two similar systems, These systems were develop ed by the same organization, and used the same development process, Fo r the environment we studied, applying principal-components analysis t o the raw data yields a neural-network model whose predictive quality is statistically better than a neural-network model developed using th e raw data alone. The improvement in model predictive quality is appre ciable from a practitioner's point of view. We concur with published L iterature regarding the number of hidden layers needed in a neural-net work model, A single hidden layer of neurons yielded a network of suff icient generality to be useful when predicting faults. This is importa nt, because networks with more hidden layers take correspondingly more time to train, The application of alternative network architectures a nd training algorithms in software engineering should continue to be i nvestigated.