You are using Amazon EC2 instances for training a machine learning model and want to ensure optimal resource utilization while maintaining cost-effectiveness. Which Amazon EC2 feature should you primarily use to automatically adjust the number of instances based on the workload during model training?