ICS Theory Group

ICS 269, Spring 1997: Theory Seminar

18 Apr 1997:
Learning Classifications via Generalized Linear Models
Jonathan Martin, ICS, UC Irvine

Though widely used in statistics and the social sciences, regression models have been overlooked or rejected by the machine learning field in favor of decision trees or neural networks. The experimental results to be reported here show that generalized linear models are an effective machine learning tool and that the classifiers inferred via these models are frequently much more accurate than those inferred by conventional decision tree methods.