Abstract:

The traditional association rules mining algorithms preferred to applying single sliding time window, and the mined alarm association rules were high support-high confidence. To solve this problem, a method was proposed to calculate comprehensive related confidence between the time adjacent items by the confidence fusion algorithm. Then generated alarm transactions were generated based on the calculated results, which enhanced alarm correlation within a transaction. The proposed approach can reduce the choice of support requirement and mine low support-high confidence association rules effectively. Based on h-confidence theory, interesting alarm association rules were screened. Experimental results showed the feasibility and effectiveness of this approach.

Key words: alarm association rule, comprehensive related confidence, h-confidence theory