UCRP-miner: Mining Patterns that Matter

Sadeq Darrab, David Broneske, Gunter Saake

Abstract: The primary goal of analyzing data is to uncover patterns that generate novel, unknown and unexpected knowledge. In data mining, frequent pattern mining is a well-known method of finding patterns that co-occur frequently. There are, however, two significant problems with frequent pattern mining, namely pattern explosion and rare (infrequent) pattern mining. Many algorithms have been proposed to extract a compact representation of patterns in data and detect interesting patterns that infrequently occur. Despite this tendency, traditional methods primarily employ frequency-driven techniques, which may not be suitable for real-world domains characterized by unusual patterns that generate unforeseen knowledge. Unexpected pattern mining is used to detect patterns in data that do not adhere to normal behavior. Recently, unsupervised clustering models have been proposed for identifying unexpected rules. These models have two significant limitations: retrieving redundant ones and failing to discover the complete set of interesting patterns. In addition, their performance is degraded in terms of time and memory. This paper introduces a novel model, UCRP-miner, to efficiently retrieve the complete set of interesting patterns. It utilizes known knowledge (frequent closed patterns) to produce new, known, and unpredictable knowledge. to evaluate the proposed model, we evaluate it using real-life datasets. The experiments show that our model generates the complete set of interesting patterns that are non-redundant and differ from the set of beliefs (frequent patterns), significantly outperforming the state-of-the-art models.