Abstract
Modern society is highly dependent on software-enabled systems, including mission and life critical systems. High profile failures of these systems damage trust in the maturity of the underlying technology and subsequently create concerns related to system safety and security. In the absence of objective methods to model the reliability of software within complex systems, decision-makers will struggle to deliver dependable and trustworthy systems. One standard definition of software reliability is the probability of failure-free software operation for a specified period of time in a specified environment. In past decades, researchers have proposed a variety of software reliability growth models (SRGM) to assess the reliability of software during phases of test and operation which often possess complicated parametric forms but disregard predictive accuracy. Moreover, most SRGM are restricted to defect discovery data, yet removal of these defects is the practical concern of software engineers. Traditional SRGM have also dedicated limited consideration of factors associated with software testing like the severity of defects. Prior efforts to model defect resolution are primarily based on systems of differential equations and queueing theory. However, these past modeling efforts offer little concrete guidance that software practitioners can relate to or use when attempting to improve their processes. To overcome the limitations noted above, this dissertation presents several modeling contributions including (i) a framework composed of several SRGM possessing a bathtub-shaped fault detection rate, stable and efficient model fitting algorithms, and assessment with a combination of predictive and information-theoretic measures to justify their increased complexity, (ii) connecting a NASA defect-tracking database to novel models of defect discovery and resolution, including differential equation-based, distributional, and Markovian models, and (iii) a defect resolution pre diction model that utilizes a SRGM incorporating covariates through the discrete Cox proportional hazard model.