Hands-On Artificial Intelligence for Beginners

Overfitting and underfitting

Overfitting is a phenomenon that happens when an algorithm learns it's training data too well to the point where it cannot accurately predict on new data. Models that overfit learn the small, intricate details of their training set and don't generalize well. For analogy, think about it as if you were learning a new language. Instead of learning the general form of the language, say Spanish, you've learned to perfect a local version of it from a remote part of South America, including all of the local slang. If you went to Spain and tried to speak that version of Spanish, you would probably get some puzzled looks from the locals! Underfitting would be exact opposite of this; you didn't study enough Spanish, and ...