Mining Fault-Tolerant Frequent Patterns in Large Databases

Abstract

In view of real world data may be interfered with noise which leads data to contain faults. Besides, we may hope that the knowledge discovered is more general and can be applied to find more interesting information. Hence, FT-Aprori was proposed for fault-tolerant data mining to discover information over large real-world data. However, FT-Apriori which generates and tests candidates based on Apriori property is not so efficient. In this paper, we develop memory-based algorithm FTP-mine which is based on the concept of pattern growth to mine fault-tolerant frequent patterns efficiently. In FTP-mine the table, STable, is designed to count the item support and FT-support of the k-length patterns which have the same prefix of length k-1. As to mining in a large database which is too large to fit in memory, FTP-mine also can be adopted by means of database partition. Since there might exist a large number of fault tolerant frequent patterns and some may be contained in others, we also focus on the finding of maximal FT-frequent patterns by extending the FTP-mine algorithm. Our study shows that FTP -mine has higher performance than FT-Apriori in various datasets. The empirical evaluations show the proposed method has good linear scalability and outperforms than FT-Apriori in various settings in the discovery of FT-frequent pattern.