In machine learning the validation phase verifies the model produced during the training phase. In supervised learning this typically involves some form of hold back of training data, say 10%, to be used for checking, with only the remaining 90% used for actual training. Normally the accuracy on this unseen test data set is lower than the data used for actual training. Validation of unsupervised learning is more difficult relying on intrinsic measures, such as the level of tightness of clusters.
Defined on page 177
Used on page 177