Genetic-based machine learning systems are competitive for pattern recognition

Abstract

During the last decade, research on GeneticBased Machine Learning has resulted in several proposals of supervised learning methodologies that use evolutionary algorithms to evolve rule-based classification models. Usually, these new GBML approaches are accompanied by little experimentation and there is a lack of comparisons among different proposals. Besides, the competitiveness of GBML systems with respect to non-evolutionary, highlyused machine learning techniques has only been partially studied. This paper reviews the state of the art in GBML, selects some of the best representatives of different families, and compares the accuracy and the interpretability of their models. The paper also analyzes the behavior of the GBML approaches with respect to some of the most influential machine learning techniques that belong to different learning paradigms such as decision trees, support vector machines, instance-based classifiers, and probabilistic classifiers. The experimental observations emphasize the suitability of GBML systems for performing classification tasks. Moreover, the analysis points out the strengths of the different systems, which can be used as recommendation guidelines on which systems should be employed depending on whether the user prefers to maximize the accuracy or the interpretability of the models.

An algorithm evaluation for discovering classification rules with gene expression programming

“…For the number of attributes, we applied the same process but also trying to maximize the distances between one range and other. Attribute ranges (4-9), (13)(14)(15)(16)(17)(18)(19)(20)(21) and were derived from these divisions and individual ranking taken from Table 5. Tables 8 and 9 have been created to provide better visualization of the count for classes and attributes respectively.…”

“…On the other hand, Orriols-Puig et al 18 presented UCS (the evolution of XCS) as the learning algorithm that resulted in the most accurate models on average. Also Orriols-Puig et al 18 conclude that GASSIST yielded competitive results in terms of accuracy. We found GASSIST, LOGIT-BOOST and UCS to provide very similar results in their behavior during our assessment of MCGEP.…”

“…Thus, rules should cooperate with each other to produce an optimal rule set jointly, and at the same time, the rule set is reduced by a token competition. A more detailed explanation of each of the first six algorithms above can be found condensed in an excellent review 18 . The following subsection, provides details of the real-world problems chosen for the experiments, the experimental tools and configurations parameters for each one, the experimental results and the statistical analysis applied to compare the results obtained.…”

“…To achieve repeatability of the experiments, we show the configurations of each problem in Tables 3 and 4. The source codes and optimum configurations for UCS, GASSIST, HIDER, SLAVE, LOGIT-BOOST and CORE were based in KEEL 46 , its original authors and in the useful review paper 18 . We used the JCLEC implementation of the Bojarczuk, Falco and Tan algorithms, with the configurations recommended by the original authors 19,27,41 respectively.…”

“…In recent years, evolutionary algorithms (EAs) have been used in classification tasks. The extensive and comprehensive scientific work on state-of-the-art and comparative studies on genetic rule-based systems (GRBS) by Fernandez et al 17 and Orriols et al 18 is a clear evidence of how useful these kinds of algorithms are. However, there are only few comparisons between GRBS and gene expression programming rule-based systems (GEPRBS) that support the competitiveness of the latter.…”