AN EXTENDED ADAPTIVE CUCKOO SEARCH TECHNIQUE FOR IMPROVED PERFORMANCE TESTING OF OBJECT-ORIENTED SOFTWARE

Abstract

Over the years, various optimization techniques for Unified Modeling Language (UML)-based testing of object-oriented software systems have been proposed. These techniques support automatic test case generation, enabling test scenarios to be identified early in the software development lifecycle. However, existing techniques predominantly generate test cases from one or two UML behavioral models, resulting in inadequate test coverage, while requiring many iterations to reach optimal solutions, leading to increased execution time and test effort. This limitation is mainly attributed to an imbalance between exploration and exploitation, leading to unstable search behavior, slow convergence, premature convergence, and sub￾optimal solutions. To address these gaps, this study proposed an Extended Adaptive Cuckoo Search Technique (EACST) for efficient and effective testing of object￾oriented software. The study aimed to determine the parameters, design, implement, and evaluate EACST for improving test coverage and fitness while reducing test suite size, test effort, execution time, and iterations. The study was guided by the pragmatism research philosophy and Design Science Research Methodology (DSRM). A Parameter Selection and Validation Framework integrating a systematic literature review (SLR), expert opinion survey, and Fuzzy Delphi Method (FDM)were used to identify and validate the EACST parameters. The SLR identified population size, maximum iterations, objective function, search space, Lévy flight, step size, abandonment probability, and local search, which were subsequently presented to software testing experts through an expert opinion survey using a semi￾structured questionnaire. The target population comprised software testing experts with relevant academic qualifications, professional experience, and software testing knowledge. Purposive sampling identified nine initial experts, followed by snowball sampling, resulting in 56 potential experts. After screening, 40 valid responses were retained for FDM analysis, which confirmed that all identified parameters satisfied the predefined acceptance criteria. EACST integrated UML behavioral models into an Integrated System Testing Graph (iSTG) using the Adaptive Weighted UML Model Integration (AW-UMI) algorithm and incorporated adaptive Lévy-flight scaling and PSO-based local search to improve the exploration–exploitation balance. The technique was implemented in Python and evaluated against ACSA, ECSA, ABC–SA, HGGA, and DLS–DE across multiple benchmark system models using parameter sensitivity, comparative, and descriptive and inferential statistical analyses. The findings showed that ω = 0.5 provided a favorable exploration–exploitation balance, resulting in stable convergence and high solution quality. EACST achieved full coverage, small test suite size, low test effort, high fitness, and competitive execution time, although other techniques performed better on specific metrics in some systems. Overall, the findings demonstrate that EACST provides an effective and efficient approach to model-based test suite optimization. The study recommends applying EACST to larger and more diverse software systems and integrating it into practical software testing environments. Future research should investigate automated parameter tuning, additional UML models, multi-objective optimization, industrial-scale implementation, fault-detection capability, and benchmarking against recent state-of-the-art techniques.