A data scientist is running an automatic model tuning job in SageMaker to optimize an XGBoost binary classifier. The tuning job launches 200 training jobs across a wide hyperparameter search space, and each job takes about 45 minutes. The team wants to reduce the overall tuning time and cost by automatically terminating training jobs that are unlikely to improve on the best objective metric seen so far, without changing the search strategy. Which action best meets this requirement?