Algorithm Comparison for Microplastic Classification: Evaluating Ensemble Models on Density and Measurement Features

Microplastic (MP) contamination poses a critical environmental threat, especially within aquatic ecosystems where MPs can infiltrate the food chain, impacting both marine life and human health. Traditional methods for MP identification and classification, such as Fourier-Transform infrared (FTIR) spectroscopy, are limited by labor-intensive procedures and the need for specialized expertise. Recent studies have highlighted the potential of machine learning (ML) and deep learning (DL) in advancing MP analysis; however, image-based methods like convolutional neural networks (CNNs) are computationally intensive. This study addresses these limitations by implementing boosting algorithms (AdaBoost, Gradient Boosting, XGBoost, LGBM, and CatBoost) that use numerical features for enhanced classification efficiency and accuracy. Feature importance analysis reveals that "Density Range" is the most significant determinant for MP categorization, while CatBoost stands out with optimal performance on limited data. Results indicate that boosting models consistently achieve high classification accuracy, suggesting that ensemble learning approaches are more effective and scalable alternatives for MP classification compared to traditional and simpler statistical models. This approach provides a robust methodological framework for improved MP monitoring and environmental management strategies.