next up previous contents
Next: Instance Based Algorithms Up: Inducers Previous: ID3MC4

Decision Tables

One of the simplest conceivable inducers. Stores a table of all instances, predicts according to the table. If an instance is not found, table-majority predicts the majority class of the table and table-no-majority returns ``unknown'' (always wrong against test-set).

When coupled with feature subset selection it provides a powerful inducer for discrete data []. If discretization is done, it is also powerful for data with continuous attributes. For example, to run discretization and feature subset selection, one can define the following options:

   setenv INDUCER disc-filter
   setenv DISCF_INDUCER FSS
   setenv DISCF_FSS_INDUCER table-majority
   setenv DATAFILE cleve
and run the Inducer utility. Consider raising the LOGLEVEL to 2 to see the progress. You can use the project utility on the final node in order to study the selected attributes in isolation.



Ronny Kohavi
Sun Oct 6 23:17:50 PDT 1996