An efficient hybrid by partitioning approach for extractingmaximal gradual patterns in large databases (MPSGrite)

Abstract

Since automatic knowledge extraction must be performed in large databases, empirical studies are alreadyshowing an explosion in the search space for generalized patterns and even more so for frequent gradual patterns.In addition to this, we also observe a generation of a very large number of relevant extracted patterns. Being facedwith this problem, many approaches have been developed, with the aim of reducing the size of the search spaceand the waiting time for detection, for end users, of relevant patterns. The objective is to make decisions or refinetheir analyses within a reasonable and realistic time frame. The gradual pattern mining algorithms common in largedatabases are CPU intensive. It is a question for us of proposing a new approach that allows an extraction ofthe maximum frequent gradual patterns based on a technique of partitioning datasets. The new technique leadsto a new, more efficient hybrid algorithm called MSPGrite. The experiments carried out on several sets of knowndatasets justify the proposed approach.