A machine learning team has an ensemble of 5 models for a critical task. They observe that each new model added to the ensemble provides a smaller accuracy boost than the last. They are now considering two options for their next development cycle, both requiring similar engineering effort:

• Option 1: Add 5 more models to the ensemble, which is projected to increase overall accuracy by 0.2% and significantly increase computational costs.
• Option 2: Focus on improving the quality of the training data for the existing 5 models, which is projected to increase overall accuracy by 1.5% without increasing computational costs.

Which option should the team choose? Justify your recommendation by explaining the underlying principle that makes one option more efficient than the other in this scenario.