NHOP: Detecting descriptive patterns using association pattern mining

To facilitate interpretation and consider the relationships of internal interdependency between data values, a new form of high-order (multiple-valued) pattern known as Nested High-Order Pattern (NHOP) is presented. This pattern satisfies a consistent statistical criterion when the pattern is iteratively extracted. The general form of High-Order Pattern (HOP), that NHOP is a subtype, is a set of multiple associated values (identified as variable outcomes) extracted from a random N-tuple. The pattern is detected by statistical testing if the occurrence is significantly deviated from the expected according to a prior model or null hypothesis. Here we extend our work of NHOP to classification and clustering tasks to identify a clearer description of the patterns. The rationale is that, complex association patterns reinforce the underlying meaningfulness in interpreting the regularity, thus can provide a better understanding of the classification and data domain. We develop a search strategy, called NHOP-Covering algorithm, to detect a set of NHOP. Experiments on benchmark machine learning data and real-world biomolecular data are evaluated with promising results.