Learning system fault diagnostic rules: A probabilistic inductive inference approach

In this paper, we present an inductive learning method that is effective for acquiring knowledge in the building of engineering expert systems. Our method is particularly suitable for the type of engineering applications that are classification in nature, such as system fault diagnosis, material and technology selection, reliability assessment, etc. The acquired classification knowledge is represented in easy-to-apply production rules that can be used to classify any new events. The performance of the method has been compared favorably elsewhere to ID3 for a number of application domains in terms of classification accuracy and computational efficiency. In this paper, we present an empirical comparison of our method to the AQ family of inductive learning systems using a typical engineering fault diagnosis task. We also suggest some reasons as to why our method performs better for some types of engineering applications, especially where the training data may be uncertain.