An efficient algorithm for discovering positive and negative patterns

In previous work we formally defined two types of potentially interesting patterns, referred to as positive and negative. These two types of patterns provide statistical knowledge that is able to affect one´s belief system. In this paper we propose two algorithms, called support-based discovery of potentially interesting patterns (SBDPIP), and all-confidence discovery of potentially interesting patterns (ACDPIP), to discover patterns that qualify as potentially interesting and compare them to discovering all potentially interesting patterns (DAPIP), an algorithm we introduced in our earlier work. ACDPIP is different from the other two algorithms in that it generates dasiafrequentpsila itemsets using an all-confidence threshold. We establish a lower bound for the threshold on all-confidence value for coverage and show empirically that ACDPIP algorithm represents an efficient alternative to DAPIP and significantly outperforms the algorithm SBDPIP with respect to coverage.