Cost-Sensitive Classification for Medical Diagnosis

Abstract

In this article, we present a cost-sensitive approach to medical diagnosis based on fuzzy rule-based classification (Schaefer, Nakashima, Yokota, & Ishibuchi, 2007). While fuzzy rule-based systems have been mainly employed for control problems (Lee, 1990) more recently they have also been applied to pattern classification problems (Ishibuchi & Nakashima, 1999; Nozaki, Ishibuchi, & Tanaka, 1996). We modify a fuzzy rule-based classifier to incorporate the concept of weight which can be considered as the cost of an input pattern being misclassified. The pattern classification problem is thus reformulated as a cost minimisation problem. Based on experimental results on the Wisconsin breast cancer dataset, we demonstrate the efficacy of our approach. We also show that the application of a learning algorithm can further improve the classification performance of our classifier.

Key Terms in this Chapter

Pattern Classific ation: Automatic transformation of input data into categories (classes).

Membership Function: A function that describes the degree of an element’s membership in a fuzzy set.

Cross Validation: A standard method of evaluating classifiers where data with known classes is divided into disjoint sets and the classifier tested on various combinations of these training sets.

Fuzzy Set: An extension of the classical set whose memberships have degrees of membership.

Fuzzy Logic: A form of logic where variables can take on variable degrees of truth.

Rule-Based Classifier: A pattern classification system in which the classification is expressed as a set of rules.