Multi-strategy improved Parrot optimization for feature engineering in machine learning

doi:10.1007/s10586-025-05613-5

Return

Multi-strategy improved Parrot optimization for feature engineering in machine learning

2025-09-22

PRE

DOI：10.1007/s10586-025-05613-5

Original

Original request for help

Save

Abstract

En 中文

Meta heuristic algorithms have become a key tool for solving complex optimization challenges in machine learning systems, especially compared to traditional optimization algorithms, they are more effective for complex optimization problems. Parrot optimization (PO) is an efficient optimization algorithm developed in recent years, which has certain advantages compared to many classical algorithms. However, its performance is still not optimal when dealing with mixed variable problems and machine learning scenarios, such as feature subset selection in imbalanced data distributions. To overcome these limitations, this paper proposes four enhancement strategies: Circle chaotic mapping strategy, novel social communication strategy, fusion ray refraction opposition-based learning (FRR-OBL), and novel reproductive behavior strategy. A Multi-Strategy Improved Parrot Optimization (MPO) was proposed based on these improvement strategies. MPO’s convergence accuracy, speed, and robustness were evaluated as follows: Using 23 benchmark functions and CEC2022, a comparative experiment was conducted on 11 popular algorithms. By comparing these 11 competitors, the advantages of MPO were demonstrated. The MPO algorithm presented in this study addresses the shortcomings associated with the PO algorithm, showing superior performance in identifying the optimal feature subset and enhancing classification accuracy in feature selection tasks.

Keywords:

Parrot optimization

Novel social communication strategy

Novel reproductive behavior strategy

Feature selection problem