RULE LEARNING

Introduction

  Rule learning is a machine learning technique that induce a target function from examples. The target function is defined jointly as a set of if-then rules.

The technique has established itself a basic component of many machine learning systems, and has been the first machine learning technology to deliver commercially successful applications (e.g. GASOIL, BMT, in process control...).

This web site only introduces an algorithm in Rule Learing family, namely CN2, in top-down approach. It contains the following documents:

Sequential Covering is the most popular algorithm that rule learning employs. It learns a single rule each time, repeats process untill the final set is formed.

CN2 is one of variation of sequantial covering to be discussed here. It learns rules for all classes, and it can handle noises.

Beam search is a popular procedure to learn one rule, employed by sequential covering and particularly CN2.

An Applet demostrating CN2 in action with a given example.


 

    Developed by Canh Hao Nguyen and Hong Chung Nguyen.
    Last updated: 04/11/2001.