Using association rules to make rule-based classifiers robust

Abstract

Rule-based classification systems have been widely used in real world applications because of the easy interpretability of rules. Many traditional rule-based classifiers prefer small rule sets to large rule sets, but small classifiers are sensitive to the missing values in unseen test data. In this paper, we present a larger classifier that is less sensitive to the missing values in unseen test data. We experimentally show that it is more accurate than some benchmark classifies when unseen test data have missing values.