90 likes | 131 Views
Incremental Mining of Association Rules. Vikram Pudi Database Systems Lab, SERC Indian Institute of Science. Underlying Problem . A, B, C D, E ( support , confidence ) where Find all rules which satisfy minsupp Find all itemsets which satisfy minsupp. Incremental Mining.
E N D
Incremental Mining of Association Rules Vikram Pudi Database Systems Lab, SERC Indian Institute of Science
Underlying Problem • A, B, C D, E (support, confidence) • where • Find all rules which satisfy minsupp • Find all itemsets which satisfy minsupp
Incremental Mining • Mining is not a one-time operation • Why not use previous mining results? • Issues • Insertions, Deletions • Skew of Increment • Change in minsupp Mining Data Business Strategy (feedback)
Fast UPdate (FUP) • An itemset that wasn’t large before, can now become large only if it is large within the increment. • Makes Multiple Passes • Doesn’t compute negative border • Doesn’t handle change in minsupp
Other Algorithms • TBAR • One pass over original database • Multiple passes over increment • Borders • One pass over original database • Two passes over increment • Explosion of candidates • Don’t handle change in minsupp
DELTA • One pass over original database • Three passes over increment • Performance close to IDEAL • Handles change in minsupp
Generalized Rules • Cereal Milk (support, confidence) • Algorithms • Cumulate • Stratify • Stratify Variants • DELTA+ • IDEAL Cereal Rice Wheat