"Advanced Software Reliability Growth Modeling: Hybrid Metaheuristic Estimation under Stochastic Testing with Dependent and Time-Varying Failure Dynamics"

In the ever-evolving landscape of software engineering, the capacity to model and predict software reliability with high precision has become indispensable to ensuring operational efficiency, minimizing risk, and enhancing user trust. Traditional Software Reliability Growth Models (SRGMs) have made significant strides in capturing fault detection trends over time. However, they often falter in environments marked by uncertainty, stochastic testing efforts, and dynamically interdependent failure processes. This study introduces a novel class of advanced SRGMs that integrate time-varying stochastic testing mechanisms with hybrid metaheuristic optimization techniques to estimate parameters under real-world uncertainty. By fusing classical Non-Homogeneous Poisson Process (NHPP) modeling frameworks with dependency-aware functions and hybridized evolutionary computation—including Modified Particle Swarm Optimization (MPSO) and Genetic Algorithms (GA)—the paper presents a powerful modeling paradigm that dynamically adapts to varied reliability trajectories. The proposed framework is validated against multiple real and synthetic datasets characterized by fluctuating test effort patterns, dependent fault behaviors, and resource constraints. Extensive simulations demonstrate that the hybrid metaheuristic techniques outperform conventional Maximum Likelihood Estimation (MLE) and standalone heuristic approaches, particularly in environments where failure intensities vary with operational profile shifts. Moreover, the model supports high-resolution estimation in cloud-based and DevOps-integrated software ecosystems where time-to-market and continuous delivery cycles place a premium on reliable analytics. The paper concludes by addressing theoretical implications, practical applications, and avenues for future research into integrating AI-driven adaptivity and deep uncertainty quantification in SRGMs.