Early assessment of an approach to determining the predictive coverage of case-based reasoning with adaptation through CARMA

Abstract

CARMA is a decision-support system for managing grasshopper infestations which uses an approach called approximate-model-based adaptation whereby case-based reasoning (CBR) provides an approximate solution and model-based reasoning adapts this approximation into a precise solution. CARMA's predictive accuracy on a set of known cases confirmed the ability of the technique. The evaluation was not expanded beyond the initial set of known cases due to the human effort involved in constructing such cases. We provide an overview of CARMA, and detail initial attempts to establish a process for the automatic evaluation of such systems in order to identify potential gaps in predictive coverage using Monte Carlo methods. We propose that any generated situation which produces large adjustments in prediction during adaptation suggests a potential gap in the predictive ability of a CBR system. This represents an extension of prior CBR work which considers only the matching stage when evaluating predictive coverage. © 2014 IEEE.