IMPROVING INTERNAL-VALUED INFERENCING WITH LIKELIHOD RATIO

Abstract

This paper point out the limitations of Interval-Valued Inferencing as a defuzzification method for inference engines based on the Bandler-Kohout subproduct. As an improvement, a measurement on the likelihood of an inference result in an acceptance/rejection band suggested. With this improvement, more meaningful results are generated from a Bandler-Kohout subproduct based inference system, especially if it is implemented as a medical decision support system. To demonstrate the capability of this improvement, an experiment with a popular dataset is carried out.