Production rule extraction
Shlomo Geva · 2002
This paper describes T-REX-an algorithm for concept learning, or rule extraction from examples, for discrete input/output mapping. The algorithm generates a production rule that is similar to that produced by C4.5. Unlike C4.5 which first generates a decision tree, and then converts it to a production rule, T-REX constructs the production rule from the outset. T-REX exhibits linear complexity in the number of training examples and facilitates an efficient implementation by the use of bitmap representation, allowing serial by word, parallel by bit operations on a single processor, anti is both vectorisable and distributable on more advanced processor architectures. Results are presented for several benchmark classification problems, the MONKS, IRIS, MUSHROOMS, and the PROMOTER data sets. T-REX compares favourably with alternative methods.