Formalising optimal feature weight setting in case based diagnosis as linear programming problems

Abstract

Many approaches to case based reasoning (CBR) exploit feature weight setting algorithms to reduce the sensitivity to distance functions. In this paper, we demonstrate that optimal feature weight setting in a special kind of CBR problems can be formalised as linear programming problems. Therefore, the optimal weight settings can be calculated in polynomial time instead of searching in exponential weight space using heuristics to get sub-optimal settings. We also demonstrate that our approach can be used to solve classification problems.