UNSUPERVISED DATA MINING BY RECURSIVE PARTITIONING
In this thesis, an experimental investigation into unsupervised database mining was conducted. A novel paradigm for autonomous mining proposed by Dr. L. J. Mazlack was tested. The idea states that increasing coherence will increase conceptual information; and this in turn will reveal previously unrecognized, useful and implicit information. [Mazlack,1996] In the experiments, different partitioning heuristics were tested: arbitrary partition, balanced partition and imbalanced partition. Their usefulness and differences in result are discussed in this thesis. To assist our partitioning heuristics, a rough set based model called Total Roughness was designed to measure the crispness of a partition. This model was used in our experiments to help choose partitioning attribute as well as perform non-scalar data clustering. The feasibility of integrating rough set theory in unsupervised partitioning is evaluated and addressed in this thesis.
School:University of Cincinnati
School Location:USA - Ohio
Source Type:Master's Thesis
Keywords:data mining recursive partitioning rough set unsupervised knowledge discovery
Date of Publication:01/01/2002