Unsupervised learning is the branch of machine learning concerned with discovering structure in input data that carries no attached labels, in contrast to supervised learning where each training instance has a target class or value. Its two canonical problems are clustering — partitioning the input space into cohesive groups of similar points while identifying outliers that conform to no group — and association rule mining, which finds frequent item sets and derives conditional dependency rules of the form "item set A implies item set B" when both A and A ∪ B are frequent. Clustering necessarily requires an inductive bias (for example an assumed cluster shape such as ellipsoids), and association rule mining is historically named "mining" rather than "learning" because it originated in the data mining tradition rather than the statistical learning one.
Unsupervised learning is the branch of machine learning concerned with discovering structure in input data that carries no attached labels, in contrast to supervised learning where each training inst…