Seeing the forest for the trees: tree-based uncertain frequent pattern mining

• Main Author: MacKinnon, Richard Kyle
• Other Authors: Leung, Carson K.-S. (Computer Science)
• Published: Springer International Publishing 2016
• Subjects: Data mining; Databases
• Online Access: http://hdl.handle.net/1993/31059

Many frequent pattern mining algorithms operate on precise data, where each data point is an exact accounting of a phenomena (e.g., I have exactly two sisters). Alas, reasoning this way is a simplification for many real world observations. Measurements, predictions, environmental factors, human error, &ct. all introduce a degree of uncertainty into the mix. Tree-based frequent pattern mining algorithms such as FP-growth are particularly efficient due to their compact in-memory representations of the input database, but their uncertain extensions can require many more tree nodes. I propose new algorithms with tightened upper bounds to expected support, Tube-S and Tube-P, which mine frequent patterns from uncertain data. Extensive experimentation and analysis on datasets with different probability distributions are undertaken that show the tightness of my bounds in different situations. === February 2016