Comparing various regression methods on ensemble strategies in differential evolution

TL;DR

This paper explores using various regression models to predict the most effective differential evolution strategy during optimization, demonstrating that random forest regression significantly improves performance on benchmark functions.

Contribution

It introduces a novel approach of applying regression models to dynamically select differential evolution strategies, enhancing optimization efficiency.

Findings

Random forest regression outperforms other methods in strategy prediction.

The proposed method improves optimization results on benchmark functions.

Dynamic strategy selection adapts better to problem characteristics.

Abstract

Differential evolution possesses a multitude of various strategies for generating new trial solutions. Unfortunately, the best strategy is not known in advance. Moreover, this strategy usually depends on the problem to be solved. This paper suggests using various regression methods (like random forest, extremely randomized trees, gradient boosting, decision trees, and a generalized linear model) on ensemble strategies in differential evolution algorithm by predicting the best differential evolution strategy during the run. Comparing the preliminary results of this algorithm by optimizing a suite of five well-known functions from literature, it was shown that using the random forest regression method substantially outperformed the results of the other regression methods.