Nondeterministic algorithms for solving the dynamic QoS-aware service composition under ambiguous QoS parameters

Abstract

This thesis addresses three pivotal challenges in web services environment : QoS-aware service composition under uncertainty, QoS prediction in dynamic environments, and a comprehensive review of collaborative filtering techniques for QoS prediction. The first contribution introduces an extended artificial bee colony algorithm with local search (EABC) to solve the interval-constrained QoS-aware service composition (IQSC) problem. Here, QoS uncertainty is modeled using an interval-number representation and the skyline operator is applied to eliminate redundant services, with experimental results demonstrating superior performance compared to skyline-based PSO, an fficient discrete gbest-guided artificial bee colony, and Harris Hawks optimization with an elite evolutionary strategy. The second contribution, denoted as QSCFIoT, tackles QoS ambiguity in fuzzy IoT environments by representing QoS parameters with generalized trapezoidal fuzzy numbers. This approach integrates a fuzzy skyline-based module with an improved discrete flower pollination algorithm, and its efficacy is confirmed through experiments on both real and synthetic datasets, outperforming EFPA, PSO, and ITL-QCA in terms of composition quality, computational time, and stability. The third contribution is a systematic literature review (SLR) that rigorously examines QoS prediction methods for web services, focusing on Collaborative Filtering (CF) techniques in static and dynamic settings. Following PRISMA guidelines, 512 studies were initially identified and 146 were thoroughly analyzed, revealing that while traditional CF methods perform well in static environments, they face significant challenges in dynamic contexts due to data sparsity and variability. This review highlights the advancements in hybrid and context-aware models and underscores the need for adaptive, real-time prediction approaches to better meet user demands. Collectively, these contributions provide a robust framework for enhancing both the composition and prediction of QoS in web services, advancing their reliability and adaptability in complex, real-world scenarios.