A local rule-based attribute weighting scheme for a case-based reasoning system for radiotherapy treatment planning

Abstract

This paper presents a novel local rule-based weighting scheme to determine attribute weights in a case-based reasoning system for radiotherapy treatment planning in brain cancer. A novel method of generating IF THEN rules to assign local weights to case attributes used in the nearest neighbour similarity measure is presented. The rules are prescreened using the data mining evaluation measures of confidence and support. Unique rules are then selected from the set of prescreened rules using an instance weighting algorithm that is based on a novel concept called the random retrieval probability of a training case, which is introduced to give an indication of the validity of the feedback obtained from a successful retrieval with respect to a particular training case. Experiments using real world brain cancer patient data show promising results. © 2012 Springer-Verlag.