A diagnostic system for the French long distance network using neural trees and a rule-based system

Building a diagnostic system for the French network is a complex pattern recognition problem. A two-level system is proposed to simplify the problem. The first level realizes local diagnosis on each exchange and the second level uses local diagnosis to make a general diagnosis concerning the entire network. Neural trees with ambiguity rejection that represent original nonparametric classifiers are used to build up the first level. A rule-based system is used to implement the second level