Robust Data Sampling in Machine Learning: A Game-Theoretic Framework for Training and Validation Data Selection