With growth in demand for zero defects, predicting reliability of software products is gaining importance. Software Reliability Growth Models (SRGM) are used to estimate the reliability of a software product. We have a large number of SRGM; however none of them works across different environments. Recently, Artificial Neural Networks have been applied in software reliability assessment and software reliability growth prediction. In most of the existing research available in the literature, it is considered that similar testing effort is required on each debugging effort. However, in practice, different amount of testing efforts may be required for detection and removal of different type of faults on basis of their complexity. Consequently, faults are classified into three categories on basis of complexity: simple, hard and complex. In this paper we apply neural network methods to build software reliability growth models (SRGM) considering faults of different complexity. Logistic learning function accounting for the expertise gained by the testing team is used for modeling the proposed model. The proposed model assumes that in the simple faults the growth in removal process is uniform whereas, for hard and complex faults, removal process follows logistic growth curve due to the fact that learning of removal team grows as testing progresses. The proposed model has been validated, evaluated and compared with other NHPP model by applying it on two failure/fault removal data sets cited from real software development projects. The results show that 113 Int. J. Rel. Qual. Saf. Eng. 2008.15:113-127. Downloaded from www.worldscientific.com by GEORGE WASHINGTON UNIVERSITY on 02/07/15. For personal use only.the proposed model with logistic function provides improved goodness-of-fit for software failure/fault removal data.

Modeling of software reliability has gained lot of importance in recent years. Use of software-critical applications has led to tremendous increase in amount of work being carried out in software reliability growth modeling. Number of analytic software reliability growth models (SRGM) exists in literature. They are based on some assumptions; however, none of them works well across different environments. The current software reliability literature is inconclusive as to which models and techniques are best, and some researchers believe that each organization needs to try several approaches to determine what works best for them. Data-driven artificial neural-network (ANN) based models, on other side, provide better software reliability estimation. In this paper we present a new dimension to build an ensemble of different ANN to improve the accuracy of estimation for complex software architectures. Model has been validated on two data sets cited from the literature. Results show fair improvement in forecasting software reliability over individual neural-network based models.

Models that describe the failure phenomenon and consequent enhancement in reliability due to fault removal are termed as Software Reliability Growth Models (SRGM). As the size of software system is large and the number of faults detected during the testing phase becomes large, so the change of the number of faults that are detected and removed through each debugging becomes sufficiently small compared with the initial fault content at the beginning of the testing phase. In such a situation, we can model the software fault detection process as a stochastic process with continuous-state space. Several continuous-state space SRGM based on stochastic differential equations of c Ito type to assess software reliability for large scale software systems have been proposed in the literature so far. However these continuous-state space SRGM have seldom taken the effect of testing-effort into consideration. The resources, such as manpower used for fault detection/removal, number of executed test-cases and CPU hours spent in executing software under test, are well-known as one of the most important factors related to the software reliability growth process. Some of the existing SRGM define errors of different severity. Severity of a failure or fault is the impact it has on the operation of a software-based system. Different faults may require different amount of testing efforts and testing strategy for their removal from the software. The aim of this paper is to determine the type of faults and their proportion present in the software. We have also assumed that learning of removal team grows as testing progresses due to experience and have incorporated the logistic removal rate during modeling of different types of faults with testing-effort by applying a mathematical technique of stochastic differential equations of c Ito type. Finally, a goodness-of-fit comparison between proposed models and existing continuous-state space SRGM using stochastic differential equations has been conducted.