A rule-based system applies explicit rules to provide stable, predictable outputs when an ML model is unavailable or performing poorly. It can also handle simple cases while the model handles more complex scenarios. section 13.4
Use a rule-based system as part of a fallback plan by defining:
- The conditions that activate it, such as model unavailability or poor performance. section 13.4
- The switching method between the ML model and rules: automatic, manual, or hybrid. section 13.4
- The boundary between simple cases handled by rules and complex cases handled by the model. section 13.4
Fallback during model failure
When the model is unavailable or performs poorly, the system can return a rule-based response instead of failing silently. A hybrid design can reserve rules for simple cases and the ML model for complex ones, though this adds complexity and breakpoints. section 13.4
