An Efficient Algorithm for Proportionally Fault-tolerant Data Mining

• Main Authors: Yu-tzu Lin, 林鈺慈
• Other Authors: Guanling Lee
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/95270366119051968719

碩士 === 國立東華大學 === 資訊工程學系 === 92 === Abstract Frequent pattern mining problem has been studying for several years, while few works discuss on fault-tolerant pattern mining. Fault-tolerant data mining extracts more interesting information from real world data which may be polluted by noise. However, those few previous works either not define the problem maturely or restrict the problem to finding those patterns tolerate fixed number of fault items. In this paper, the problem of mining proportionally fault-tolerant frequent patterns is discussed. Moreover, two algorithms are proposed to solve it. The first algorithm, named FT-BottomUp, applies FT-Apriori heuristic and performs the idea of finding all FT-patterns with all possible number of faults. The second algorithm, FT-LevelWise, divides all FT-patterns into several groups by their number of tolerable faults, and mines the content patterns of each group respectively. The experiment result shows more potential fault-tolerant patterns are extracted by our approach. Our contribution is offering a different type of fault-tolerant frequent pattern, in those patterns, the number of tolerated faults is proportional to the length of patterns. This gives the user another choice when traditional fault-tolerant frequent pattern mining result can’t satisfy them.