In the field of Search-Based Software Testing (SBST), effectively managing the amount of time spent on searching is essential for the automated development of test cases to be successful. Current approaches, which rely on heuristic procedures or single prediction models, often prove inadequate. This study presents the EP-BAST approach, which utilizes ensemble machine learning models to optimize the allocation of search time budget in SBST (Software-Based Software Testing). We conduct a thorough assessment of EP-BAST by examining four open-source projects from Defects4J to evaluate its efficacy against the commons use of single prediction models. The findings of our study demonstrate that EP-BAST outperforms single-model methods, enhancing the efficiency and efficacy of automatically generated test cases by using a broader range of prediction models. This work emphasizes the benefits of using ensemble machine learning models in SBST, demonstrating a more adaptable, resilient, and effective method for improving test case generation. The findings from EP-BAST lay the foundation for future advancements in automated software testing driven by machine learning.