Exploring and challenging the conventional wisdom of “more data → better performance” by experimenting with the interactions between sample size, attribute set, and model complexity.