| The problem of developing reliable software at a low cost and efficient performance still presents a great challenge for a software designer. To develop a reliable software system, several issues need to be addressed. These issues include the definition of reliable software, reliable development methodologies, testing methods for reliability, reliability growth prediction modeling, and accurate estimation of reliability.; Several software reliability growth models (SRGM) were proposed with the goal to estimate the number of residual software faults, which occur in the software testing process. These models have a set of parameters that need to be identified. These parameters are usually estimated using observed failure data. These failure data usually present a problem for a system designer due to the lack of measurements or outliers. Calculus-based estimation techniques like maximum likelihood and sum-square estimation were applied to the parameter estimation problem with some success, but with many restrictions to software reliability growth models. To maximize the likelihood function or minimize the sum square error function, for example, the continuity and the existence of derivatives of the evaluation function are required. These types of assumptions and required restrictions present a major problem for the development of exact models.; In this dissertation, we explore an alternative to the above approach through the usage of two types of neural networks (NN) models, the feedforward and the Radial basis function. Also, we explore the use of fuzzy rules. The problem of building a black box model for software reliability growth prediction using neural network and fuzzy logic is fully addressed. NNs have been used both to estimate parameters of a formal model and to learn to emulate the process model itself to predict future faults. Feedforward and Radial basis function have been successfully used to solve a variety of prediction problems, which include real-time control, military, and operating system applications. A set of fuzzy rules were also developed to model the dynamics of the software reliability growth models in various applications. The reported results using neural networks and fuzzy logic can improve the software reliability growth modeling solution. |
