RACER: accurate and efficient classification based on rule aggregation approach

摘要

Rule-based classification is one of the most important topics in the field of data mining due to its wide applications. This article presents a novel rule-based classifier called RACER (Rule Aggregating ClassifiER) to improve the accuracy of data classification. RACER uses a specific rule representation that enables it to consider each instance in the training data as an initial rule, without spending any cost. In order to retrieve an applicable rule set, RACER tries to combine the initial rules together. If the combined rule has a better fitness value in comparison with the two input rules, RACER combines them together. We have used seventeen different datasets from UCI machine learning database repository to evaluate RACER's capability in classifying various kinds of databases. Moreover, to assess RACER's performance, we compared our results with some other well-known classifiers including CN.2, PART, C4.5 and SVM. Our experiments show that RACER is an effective classifier in various domains and has better average classification accuracy and understandability in comparison with other applied classifiers.