A Hybrid Competitive Swarm Optimization Approach for High-Dimensional Feature Selection in Facial Beauty Analysis

The study of facial beauty has attracted increasing attention from researchers and has achieved notable results. However, traditional machine learning approaches to facial beauty analysis focused on limited geometric features, often fewer than 100, constrained by subjective interpretations. To capture more comprehensive geometric features and reduce reliance on subjectivity, this study introduces an enhanced high-dimensional geometric feature set with 2928 dimensions, comprising distance, angle, and ratio features. To address the challenge of redundant and irrelevant features in the high-dimensional data, this study proposes a novel hybrid competitive swarm optimization (CSO) algorithm which has the capability in handling high-dimensional data optimization. However, the traditional CSO algorithms face issues such as a lack of strong directional guidance. Our improvement integrates filter and wrapper feature selection methods, using feature importance for population initialization and dynamic embedding of feature weights during the search process. The algorithm enhances diversity and correlation by improving the particle update strategy, increasing group particle numbers, and utilizing successful particles for updates. Experiments conducted on the SCUT-FBP5500 and SCUT-FBP datasets demonstrate the effectiveness of the proposed hybrid approach in optimizing high-dimensional feature selection, providing a robust framework for analyzing facial beauty.