Genetic Programming Symbolic Classification: A Study

Created by W.Langdon from gp-bibliography.bib Revision:1.8051

@InProceedings{Korns:2017:GPTP,

• author = "Michael F. Korns",
• title = "Genetic Programming Symbolic Classification: A Study",
• booktitle = "Genetic Programming Theory and Practice XV",
• editor = "Wolfgang Banzhaf and Randal S. Olson and William Tozier and Rick Riolo",
• year = "2017",
• series = "Genetic and Evolutionary Computation",
• pages = "39--54",
• address = "University of Michigan in Ann Arbor, USA",
• month = may # " 18--20",
• organisation = "the Center for the Study of Complex Systems",
• publisher = "Springer",
• keywords = "genetic algorithms, genetic programming",
• isbn13 = "978-3-319-90511-2",
• URL = "https://link.springer.com/chapter/10.1007/978-3-319-90512-9_3",
• DOI = "doi:10.1007/978-3-319-90512-9_3",
• abstract = "While Symbolic Regression (SR) is a well-known offshoot of Genetic Programming, Symbolic Classification (SC), by comparison, has received only meagre attention. Clearly, regression is only half of the solution. Classification also plays an important role in any well rounded predictive analysis tool kit. In several recent papers, SR algorithms are developed which move SR into the ranks of extreme accuracy. In an additional set of papers algorithms are developed designed to push SC to the level of basic classification accuracy competitive with existing commercially available classification tools. This paper is a simple study of four proposed SC algorithms and five well-known commercially available classification algorithms to determine just where SC now ranks in competitive comparison. The four SC algorithms are: simple genetic programming using argmax referred to herein as (AMAXSC); the M2GP algorithm; the MDC algorithm, and Linear Discriminant Analysis (LDA). The five commercially available classification algorithms are available in the KNIME system, and are as follows: Decision Tree Learner (DTL); Gradient Boosted Trees Learner (GBTL); Multiple Layer Perceptron Learner (MLP); Random Forest Learner (RFL); and Tree Ensemble Learner (TEL). A set of ten artificial classification problems are constructed with no noise. The simple formulas for these ten artificial problems are listed herein. The problems vary from linear to nonlinear multimodal and from 25 to 1000 columns. All problems have 5000 training points and a separate 5000 testing points. The scores, on the out of sample testing data, for each of the nine classification algorithms are published herein.",
• notes = "GPTP 2017, Part of \cite{Banzhaf:2017:GPTP} published after the workshop in 2018",

}

Genetic Programming entries for Michael Korns

Citations