A paradigm for detecting cycles in large data sets via fuzzy mining

Traditional data mining algorithms identify associations in data that are not explicit. Cycle mining algorithms identify meta-patterns of these associations depicting inferences forming chains of positive and negative rule dependencies. This paper describes a formal paradigm for cycle mining using fuzzy techniques. To handle cycle mining of large data sets, which are inherently noisy, we present the /spl alpha/-cycle and /spl beta/-cycle, the underlying formalism of the paradigm. Specifically, we show how /spl alpha/-cycles, desirable cycles, can be reinforced such that complete positive cycles are created, and how /spl beta/-cycles can be identified and weakened. To accomplish this, we introduce the concept of /spl Omega/ nodes that employ an alterability quantification, as well as using standard rule and node weighting (with associated thresholds).