The Use of Learning Curves for Machine Learning Diagnostic in Cloud Resource Optimization - FittedCloud

When we use the 1st, 2nd, and nth (with a large n) order polynomial regression, we may get the following results. As we can see, the 1st order model is too simple to fit the data, the 2nd order model has reasonable fitting, and the nth order model is very complicated and can fit every single sample very well. But the problem with the 3rd case is that machine learning is not about fitting the training data, but about generalizing beyond the training data. While it can achieve small training error, it will have large test error.