Mining typical patterns from databases

There have been many approaches used to discover useful information patterns from databases, such as concept description, associations, sequential patterns, classification, clustering, and deviation detection. This paper proposes a new type of information pattern, called a typical pattern, which is a small subset of objects selected from a large dataset that provides a compact and suitable representation of the original dataset. The Typical Patterns Mining (TPM) algorithm is developed to mine typical patterns from databases. Extensive experiments are carried out using a real dataset to demonstrate the usefulness of typical patterns in practical situations. The experimental results indicate that TPM is a computationally efficient method and that typical patterns can provide a compact and suitable representation of the original dataset.