Evolutionary data mining approaches for rule-based and tree-based classifiers
Thomas Weis, Raymond Chiong · 2010
Data mining is an important process, with applications found in many business, science and industrial problems. While a wide variety of algorithms have already been proposed in the literature for classification tasks in large data sets, and the majority of them have been proven to be very effective, not all of them are flexible and easily extensible. In this paper, we introduce two new approaches for synthesizing classifiers with Evolutionary Algorithms (EAs) in supervised data mining scenarios. The first method is based on encoding rule sets with bit string genomes and the second one utilizes Genetic Programming to create decision trees with arbitrary expressions attached to the nodes. Comparisons with some sophisticated standard approaches, such as C4.5 and Random-Forest, show that the performance of the evolved classifiers can be very competitive. We further demonstrate that both proposed approaches work well across different configurations of the EAs.