On Selection of a Benchmark by Determining the Algorithms' Qualities

The authors got the motivation for writing the article based on an issue, with which developers of the newly developed nature-inspired algorithms are usually confronted today: How to select the test benchmark such that it highlights the quality of the developed algorithm most fairly? In line with this, the CEC Competitions on Real-Parameter Single-Objective Optimization benchmarks that were issued several times in the last decade, serve as a testbed for evaluating the collection of nature-inspired algorithms selected in our study. Indeed, this article addresses two research questions: (1) How the selected benchmark affects the ranking of the particular algorithm, and (2) If it is possible to find the best algorithm capable of outperforming all the others on all the selected benchmarks. Ten outstanding algorithms (also winners of particular competitions) from different periods in the last decade were collected and applied to benchmarks issued during the same time period. A comparative analysis showed that there is a strong correlation between the rankings of the algorithms and the benchmarks used, although some deviations arose in ranking the best algorithms. The possible reasons for these deviations were exposed and commented on.

Automated summary

This article discusses how to fairly evaluate the quality of nature-inspired algorithms by selecting test benchmarks and the correlation between algorithm rankings and different benchmarks. The study shows that the selected benchmark can affect the ranking of a particular algorithm, leading to deviations in the order of best-performing algorithms on different benchmarks.