Mining data stream

CREATED: 200906100345 Speaker: Philip S. Yu, UI Chicago ** Applications

• surveillance/monitoring
• click stream mining ** Issues
• real time, one pass
• resource constraints
• evolving stream characteristics, temporal locality, outliers
• noisy data ** Stream clustering
• uncertain data
• CluStream ** offline: generate stats ** online: use stats to answer query
• Statistics to generate ** micro clusters, temporal extension to CF tree ** Nearest neighbor used for update ** Stream Classification
• high dimension
• locality based: NN, decision tree, rules (DT and rules are subspace methods)
• how to choose subspace? ** sampling ** binning/histogram