On the Best Search Method in the LEM1 and LEM2 Algorithms
Jerzy W. Grzymala‐Busse, Paolo Werbrouck · Studies in fuzziness and soft computing · 1998
This report presents results of experiments on two algorithms of machine learning: LEM1 and LEM2. Both algorithms belong to the LEM (Learning from Examples Module) family developed at the Department of Computer Science, University of Kansas. For LEM1, two different approaches to test attribute dependence were compared: partition and lower boundaries. The two different versions of the algorithm were developed and their run times on a set of test files were compared. For LEM2 a number of experiments were made to find the best search method of the description space. Some heuristics used within the algorithm to pick the “best” attribute-value pairs for the generation of rules were selected and tested. The quality of different methods has been compared on the basis of the total number of conditions, the total number of rules, and the average length of rules. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.