Evaluating the Performance of an Evolutionary Tool for Exploring Solution Fronts

Urquhart, Neil

Abstract

EvoFilter is an evolutionary algorithm based tool for searching through large non-dominated fronts in order to find a subset of solutions that are of interest to the user. EvoFilter is designed to take the output of existing Multi Objective Evolutionary Algorithms and act as a decision support tool for users. Currently EvoFilter is available for all to use on-line \cite{urquhart-2017b}. This paper evaluates the performance of EvoFilter by creating a large number of randomised filter specifications which are then applied using EvoFilter and a simple filter to a range of non-dominated fronts created by a portfolio of Multi Objective Genetic Algorithms (MOGAs). The results show that EvoFilter is capable of finding sets of solutions that meet the users' requirements more closely than those found using the simple filter. EvoFilter increases performance on some objectives by including relevant solutions event if these solutions slightly lessen performance on other objectives. The filter discussed in this paper may be accessed online.

Citation

Urquhart, N. (2018, April). Evaluating the Performance of an Evolutionary Tool for Exploring Solution Fronts. Presented at EvoApplications 2018

• Presentation Conference Type: Conference Paper (published)
• Conference Name: EvoApplications 2018
• Start Date: Apr 4, 2018
• End Date: Apr 6, 2018
• Acceptance Date: Jan 3, 2018
• Online Publication Date: Mar 8, 2018
• Publication Date: Mar 8, 2018
• Deposit Date: Apr 24, 2018
• Publicly Available Date: May 8, 2018
• Publisher: Springer
• Pages: 523-537
• Series Title: Lecture Notes in Computer Science
• Series Number: 10784
• Series ISSN: 0302-9743
• Book Title: Applications of Evolutionary Computation
• Chapter Number: 36
• ISBN: 978-3-319-77538-8
• DOI: https://doi.org/10.1007/978-3-319-77538-8_36
• Keywords: Visualisation, multi-objective, optimisation, scheduling,
• Public URL: http://researchrepository.napier.ac.uk/Output/1023121
• Contract Date: Apr 24, 2018