You are working on a large-scale machine learning project where you are using TensorFlow to train a model on a structured dataset containing 100 billion records. These records are currently stored in multiple CSV files. To optimize the input/output execution performance and ensure efficient data processing and training, what should you do?