A Comparison of Data Sources forMachine Learning in a Telephone TroubleScreening Expert System

This paper describes a domain where the application of machine learning, speciically inductive learning, could have enormous positive impact. The domain possesses attributes that would indicate that in-ductive learning would easily succeed for this domain. In particular, data for this domain are abundant. In spite of this, numerous machine learning methods { both inductive and otherwise { have failed to learn a knowledge base having high accuracy. This paper presents a comparison of the data sources available for this domain. It focuses primarily on a survey system that was ultimately designed for the purpose of collecting data best suited to this task.