An Extended Adaptive Cuckoo Search Technique for Test Case Optimization

Model-driven test case production using Unified Modeling Language (UML) models has been widely explored to support early testing, often before system implementation. However, many existing techniques still struggle to produce optimal results efficiently, requiring a large number of iterations while achieving limited test coverage, as they rely on only a few UML models. To address these limitations, this study proposes an Extended Adaptive Cuckoo Search Technique (EACST), which builds and improves the Adaptive Cuckoo Search Algorithm (ACSA). Guided by the Design Science Research Methodology, the technique was conceptualized, developed, and evaluated through comparative experiments involving the Adaptive Cuckoo Search Algorithm (ACSA), the Enhanced Cuckoo Search Algorithm (ECSA), the hybrid Artificial Bee Colony and Simulated Annealing algorithm (ABC_SA), the Hybrid Graph-based Genetic Algorithm (HGGA), and the Differential Evolution algorithm with exponential crossover based on a learning strategy within the difference vector (DLS_DE). The initial evaluation was conducted using a single dataset to assess baseline performance under controlled conditions. The results demonstrate that EACST achieves optimal solutions using fewer iterations and reduced execution time compared to baseline techniques. Statistical validation using the Friedman and Wilcoxon tests further confirms these improvements, showing that EACST significantly outperforms ACSA, ECSA, ABC_SA, HGGA and DLS_DE. To ensure generalizability, a cross-system evaluation involving ten datasets was subsequently performed. The results show that EACST consistently generates a minimal test suite with maximum coverage, reduced effort, and fewer iterations compared to existing techniques. The proposed technique can be applied to a wide range of systems, including real-time environments such as Internet of Things systems and autonomous vehicles that utilize object-oriented paradigms. Future work will focus on developing hybrid approaches by integrating the proposed technique with Large Language Models and Reinforcement Learning to further enhance performance.