GRACER: Improving the Accuracy of RACER Classifier Using A Greedy Approach

Rule-based classifiers play a prominent role in the field of data mining as they are used frequently in tackling challenges from medical to industrial applications. RACER (Rule Aggregating ClassifiER), a recently-introduced algorithm, is one of the rule-based classifiers that has had increased the accuracy of classification compared to some other well-known classifiers. RACER considers each instance in the training data as an initial rule. To form an applicable rule-set, however, RACER tries to merge the initial rules and replaces the merged rule with the two initial rules whenever it has a better fitness value than the initial rules. In this paper, we have changed the rule-merging phase of the RACER classifier using a greedy approach and have proposed a new classifier based on this modification called, GRACER (Greedy RACER). In order to approve the GRACER capability, six datasets from the UCI machine learning database repository have been applied. The performance of GRACER has been compared with that of RACER and some other well-known classifiers. Our experiments show the supenority of GRACER compared with other applied classifiers.