Genetic rule extraction optimizing brier score

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

@InProceedings{Johansson:2010:gecco,

• author = "Ulf Johansson and Rikard Konig and Lars Niklasson",
• title = "Genetic rule extraction optimizing brier score",
• booktitle = "GECCO '10: Proceedings of the 12th annual conference on Genetic and evolutionary computation",
• year = "2010",
• editor = "Juergen Branke and Martin Pelikan and Enrique Alba and Dirk V. Arnold and Josh Bongard and Anthony Brabazon and Juergen Branke and Martin V. Butz and Jeff Clune and Myra Cohen and Kalyanmoy Deb and Andries P Engelbrecht and Natalio Krasnogor and Julian F. Miller and Michael O'Neill and Kumara Sastry and Dirk Thierens and Jano {van Hemert} and Leonardo Vanneschi and Carsten Witt",
• isbn13 = "978-1-4503-0072-8",
• pages = "1007--1014",
• keywords = "genetic algorithms, genetic programming, Genetics based machine learning",
• month = "7-11 " # jul,
• organisation = "SIGEVO",
• address = "Portland, Oregon, USA",
• DOI = "doi:10.1145/1830483.1830668",
• publisher = "ACM",
• publisher_address = "New York, NY, USA",
• abstract = "Most highly accurate predictive modelling techniques produce opaque models. When comprehensible models are required, rule extraction is sometimes used to generate a transparent model, based on the opaque. Naturally, the extracted model should be as similar as possible to the opaque. This criterion, called fidelity, is therefore a key part of the optimisation function in most rule extracting algorithms. To the best of our knowledge, all existing rule extraction algorithms targeting fidelity use 0/1 fidelity, i.e., maximise the number of identical classifications. In this paper, we suggests and evaluate a rule extraction algorithm using a more informed fidelity criterion. More specifically, the novel algorithms, which is based on genetic programming, minimises the difference in probability estimates between the extracted and the opaque models, by using the generalised Brier score as fitness function. Experimental results from 26 UCI data sets show that the suggested algorithm obtained considerably higher accuracy and significantly better AUC than both the exact same rule extraction algorithm maximizing 0/1 fidelity, and the standard tree inducer J48. Somewhat surprisingly, rule extraction using the more informed fidelity metric normally resulted in less complex models, making sure that the improved predictive performance was not achieved on the expense of comprehensibility.",
• notes = "Also known as \cite{1830668} GECCO-2010 A joint meeting of the nineteenth international conference on genetic algorithms (ICGA-2010) and the fifteenth annual genetic programming conference (GP-2010)",

}

Genetic Programming entries for Ulf Johansson Rikard Konig Lars Niklasson

Citations