Towards Model Selection using Learning Curve Cross-Validation