Evaluating the Impact of Particle Swarm Optimization Based Feature Selection on Support Vector Machine Performance in Coral Reef Health Classification

Abstract

This research explores improving coral reef image classification accuracy by combining Histogram of Oriented Gradients (HOG) feature extraction, image classification with Support Vector Machine (SVM), and feature selection with Particle Swarm Optimization (PSO). Given the ecological importance of coral reefs and the threats they face, accurate classification of coral reef health is essential for conservation efforts. This study used healthy, whitish, and dead coral reef datasets divided into training, validation, and test data. The proposed approach successfully improved the classification accuracy significantly, reaching 85.44% with the SVM model optimized by PSO, compared to 79.11% in the original SVM model. PSO not only improves accuracy but also reduces running time, demonstrating its effectiveness and computational efficiency. The results of this study highlight the potential of PSO in optimizing machine learning models, especially in complex image classification tasks. While the results obtained are promising, the study acknowledges several limitations, including the need for further validation with larger and more diverse datasets to ensure model robustness and generalizability. This research contributes to the field of marine ecology by providing a more accurate and efficient coral reef classification method, which can be applied to other image classifications.