Towards Efficient Discovery of Frequent Patterns with Relative Support

Frequent patterns are an important class of regularities that exist in a database. Although there exists no universally acceptable best measure to assess the interestingness of a pattern, relative support is emerging as a popular measure to discover frequent patterns involving both frequent and rare items. An Apriori-like algorithm known as Relative Support Apriori (RSA) has been discussed in the literature to discover the patterns. It has been observed that mining frequent patterns with RSA is a computationally expensive process because the discovered patterns do not satisfy the anti-monotonic property. Moreover, RSA also suffers from the performance problems involving generation of the huge number of candidate patterns and multiple scans on the database. This paper makes an effort to discover frequent patterns effectively with the relative support measure. To reduce the computational cost, we theoretically show that the patterns discovered with the relative support measure satisfy the convertible anti-monotonic property. Using this property, a pattern-growth algorithm known as Relative Support Frequent Pattern-growth (RSFP-growth) has been proposed to discover the patterns. Experimental results on both synthetic and real-world datasets show that the proposed RSFP-growth algorithm is significantly better than the RSA algorithm.