A proposed framework for the analysis of the performance of newly proposed metaheuristics

Abstract

Metaheuristics benchmarking plays a key role in developing new algorithms for optimization problems. However, a number of published studies criticize the lack of reliable and repeatable experimentation in the analysis of many newly proposed metaheuristics. The enhancement of an analysis framework is studied and implemented as an adequate response to this issue. This research presents a framework for the design and analysis of the performance of newly proposed metaheuristic algorithms named Metaheuristics Design and Analysis Framework (MDAF). The importance of well-constructed and controlled studies is recognized as a necessary step for the benchmarking results to be reliable and repeatable. Methods such as the classification of problem instances into categories based on a representation calculated from the FLACCO (Feature-Based Landscape Analysis of Continuous and Constrained Optimization) library are discussed. The selection of benchmarking parameters, problem instances, and statistical methods are also presented. It is observed from the analyses that the implementation of valid experimental methods is an effective strategy for Benchmarking the performance of optimization algorithms.

Date	21 Mar 2022
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Alain April (Supervisor)