Implementation and Deployment of an RDR-Based System
Paul Compton, Byeong Ho Kang · 2021
This chapter discusses practical issues in developing a Ripple-Down Rule (RDR) application. The case monitoring required by RDR provides a conservative estimate of the error rate and is a better way to validate than initial testing on unseen data. It allows the user to decide when some conclusions can be auto-validated, while others should still be monitored. The role of the expert in building an RDR system as part of their normal duties is discussed. It is suggested that the simplest approach to cornerstone cases is to check all cases for which rules were added against the new rule. Even with thousands of cases the user is likely to have to consider only two or three. Although RDR systems do not require an explanation capability when building rules, RDR, like case-based reasoning provides a superior explanation by providing the cases for which rules were added. If browsing and exploring the knowledge base is desired, formal concept analysis is a good starting point. Finally, the chapter discusses implementation issues such as user and system interfaces, and emphasises that building the actual RDR engine is probably the simplest part of a project