Supervised Machine Learning with Python

Strategies for handling high variance

If you, instead, find yourself facing a high variance problem, we've already seen how more training data can help, to an extent. You can also perform some feature selection to pare down the model's complexity. The most robust solution lies in bagging or ensembling, which combines the output to mini models, which all, in turn, vote on each sample's label or output regression score:

In the next section, we're going to more formally define non-parametric learning algorithms and introduce decision trees.