Constructive Induction of Cartesian Product Attributes
Overview
Cartesian Product Attributes
Bayesian Classifier
Domains
Bayesian Classifiers
Learning A1 xor A3
Joining Attributes
Searching for Dependencies Backward Elimination and Joining
Results on Naive Bayesian Classifier
Learning Parity (Total of 10 features)
Alternative Strategies for Deleting and Joining Attributes
PEBLS
Results on PEBLS
Learning a Decision Tree
Results with Decision Tree Learner
Efficiency of Leave-one-out CV
Algorithm specific results
Why not join the most dependent attributes?
Entropy Results
Which joins correlate best with test set accuracy
Joining Attributes: A restricted form of Bayesian Network
Conclusions
Email: pazzani@ics.uci.edu
Home Page: http://www.ics.uci.edu/~pazzani