Parameter Tuning for the MAX Expert System

Christopher J. Merz, Missouri University of Science and Technology
M. J. Pazzani

This document has been relocated to

There were 9 downloads as of 28 Jun 2016.


We investigate methods for tuning numeric parameters in Nynex MAX, a telephone trouble screening expert system. Steepest descent, hillclimbing, and simulated annealing parameter adjustment strategies are applied to the problems of maximizing classification accuracy and minimizing misclassification cost. For both of those optimization problems we evaluate each algorithm''s ability to tune initial parameters for several situations