On local and global feature weight discovery for case-based reasoning.
Werner Dubitzky, Francisco J Azuaje, Philippe Lopes, Paul J. McCullagh, Yuebin Song · Computers and Their Applications · 1999
This paper proposes a case feature weight learning model for case-based reasoning that automates the difficult and time-consuming case engineering task of defining effective feature weights. The proposed model covers the discovery or learning of both global and local feature weights from data. It is based on the concept of introspective learning and the methods from genetic algorithms and genetic or evolution programming. The approach has been applied and evaluated on a real-world medical prognosis task. The results indicate that the global-weights learning model is both more effective (performance) and efficient (time complexity) than its local-weights counterpart.