Cumulative fast update algorithm for mining association rules

Abstract EN

This paper proposes a new candidate generation model for association rules mining based on cumulative database processing.

This model relies on successive follow up for differential rules in a way to gradually reduce the scanning needed for finding out the association rules.

This follow up is implemented by frequency buffering and means for controlling updates from one run to another, which were not maintained in FUP2 algorithm.

The algorithm proves that after buffering most of the generated rules, there might no need to scan the original database and hence an optimal performance achieved.

In addition, the proposed algorithm has contributed in association rules categorization and correspondence.

Experimental comparisons have been conducted against FUP2 and DELTA with different support and different incremental database volumes.

Results showed better evaluation relative to DELTA, and dramatically increasing performance with respect to FUP2.