On Mining Proportional Fault-Tolerant Frequent Itemsets

Mining robust frequent itemsets has attracted much attention due to its wide applications in noisy data. In this paper, we study the problem of mining proportional fault-tolerant frequent itemsets in a large transactional database. A fault-tolerant frequent itemset allows a small amount of errors in each item and each supporting transaction. This problem is challenging since the anti-monotone property does not hold for candidate generation and the problem of fault-tolerant support counting is known to be NP-hard. We propose techniques that substantially speed up the state-of-the-art algorithm for the problem. We also develop an efficient heuristic method to solve an approximation version of the problem. Our experimental results show that the proposed speedup techniques are effective. In addition, our heuristic algorithm is much faster than the exact algorithms while the error is acceptable.