<p>White-box test data generation typically relies on an optimized search through the program input space. Metaheuristic algorithms, such as genetic algorithms, particle swarm optimization, and simulated annealing, are commonly utilized to address this problem. However, it is observed that existing algorithms often fall short in generating diverse test data. Their primary focus is identifying the optimal solution rather than a diverse set of reasonable solutions. This paper aims to address the issue of limited diversity in test data generation by proposing a modified version of the pelican optimization algorithm (POA). The goal is to improve coverage and reduce the fitness evaluations required for generating test data. This study tackles the challenge of minimizing test data volume while achieving high coverage, a significant concern in automatic test data generation. The proposed approach introduces the improved POA to solve the diversity problem in test data generation. The modified algorithm outperforms eight well-known metaheuristic algorithms regarding coverage and the number of fitness evaluations needed. The approach also incorporates techniques to address the challenge of reducing test data volume while maintaining high coverage. Compared to six well-known metaheuristic algorithms, the improved POA demonstrates superior path coverage and efficiency performance. For example, when generating 1000 test cases on benchmark programs such as QuadEq, Gcd, and Bessj, our method achieved up to an 83% increase in path coverage relative to the weakest-performing baseline. This improvement is primarily due to the enhanced diversity of test data, which reduces redundancy and increases the likelihood of covering unique execution paths. The consistent performance of the proposed approach across multiple benchmarks highlights its effectiveness in generating diverse, high-coverage test suites for white-box testing.</p>