Abstract
Historically, numerical algorithms such as Newton's method were employed, which required good initial parameter estimates and therefore a high level of expertise to apply SRGM. Numerical and statistical methods for software reliability model fitting include the expectation-maximization algorithm, expectation conditional maximization algorithm, and Newton-Raphson method. When initial parameter estimates are far from the maximum, the expectation-maximization and expectation conditional maximization algorithms are stable, but can exhibit slow convergence, whereas the Newton-Raphson method often diverges. Applications of swarm intelligence algorithms to software reliability include particle swarm optimization, artificial bee colony, ant colony optimization, cuckoo search, grey wolf optimization, firefly, ant lion optimization, and whale optimization. The chapter reviews software reliability growth models and describes parameter estimation methods. It presents a sequence of illustrative examples demonstrating how particle swarm optimization is combined with traditional methods in a manner to conduct rigorous tradeoff assessment between convergence and runtime.