Formalizing Expert Knowledge Through Machine Learning

Abstract

This chapter addresses one of the key questions in service science: how to formalize expert knowledge. While this question has been treated mainly as a task of formal language design, we use an alternative approach based on machine learning. Investigating the history of expert systems in artificial intelligence, we suggest that three criteria, generalizability, learnability, and actionability, are critical for extracted expert rules. We then conclude that machine learning is a promising tool to satisfy these criteria. As a real example, we perform a case study on a task of condition-based maintenance in the railway industry. We demonstrate that our proposed statistical outlier detection method achieves good performance for early anomaly detection in wheel axles, and thus in encoding expert knowledge.