Machine learning-driven microplastics identification using ensemble stacking with Extra Tree meta-models from FTIR data

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60076658%3A12520%2F25%3A43910036" target="_blank" >RIV/60076658:12520/25:43910036 - isvavai.cz</a>

Výsledek na webu

<a href="https://doi.org/10.1016/j.jece.2025.118315" target="_blank" >https://doi.org/10.1016/j.jece.2025.118315</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.jece.2025.118315" target="_blank" >10.1016/j.jece.2025.118315</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Machine learning-driven microplastics identification using ensemble stacking with Extra Tree meta-models from FTIR data

Popis výsledku v původním jazyce

Microplastics (MPs) have become a major global environmental issue in recent decades due to their widespread presence in oceans, bioavailability, and ability to carry toxic chemicals. Attenuated Total Reflectance Fourier-Transform Infrared (ATR-FTIR) spectroscopy is widely used for MPs identification and analysis, but it faces challenges such as class imbalance, spectral similarities, and fouling, which hinder data accuracy and classification. This study proposes MLStackXT, a stacking-based machine learning (ML) model designed to enhance MPs classification by addressing these challenges using FTIR spectral datasets. The model was trained and validated using the Kedzierski and Jung dataset. It demonstrated superior performance compared to previous ML and deep learning (DL) approaches, including support vector machines (SVMs), ensemble learning, and deep neural networks (DNNs). On the Kedzierski dataset, MLStackXT achieved 95.85% accuracy, with a kappa score of 94.96%, F1-score of 95.73%, recall of 95.85%, and precision of 96.10%. Similarly, on the Jung dataset, the model attained 95.00% accuracy, a kappa score of 91.28%, an F1-score of 94.78%, a recall of 95.00%, and a precision of 94.81%. Confusion matrix analysis confirmed a significant reduction in misclassification, achieving 100% accuracy in 8 out of 12 MPs categories (Kedzierski) and over 97% accuracy in 4 out of 5 MPs categories (Jung). Model optimization via Optuna and interpretability analysis using SHAP highlighted the influence of principal component analysis (PCA) and key spectral features on classification performance. In addition, MLStackXT outperformed various stacking configurations, demonstrating its robustness and effectiveness for MPs detection.

Název v anglickém jazyce

Machine learning-driven microplastics identification using ensemble stacking with Extra Tree meta-models from FTIR data

Popis výsledku anglicky

Microplastics (MPs) have become a major global environmental issue in recent decades due to their widespread presence in oceans, bioavailability, and ability to carry toxic chemicals. Attenuated Total Reflectance Fourier-Transform Infrared (ATR-FTIR) spectroscopy is widely used for MPs identification and analysis, but it faces challenges such as class imbalance, spectral similarities, and fouling, which hinder data accuracy and classification. This study proposes MLStackXT, a stacking-based machine learning (ML) model designed to enhance MPs classification by addressing these challenges using FTIR spectral datasets. The model was trained and validated using the Kedzierski and Jung dataset. It demonstrated superior performance compared to previous ML and deep learning (DL) approaches, including support vector machines (SVMs), ensemble learning, and deep neural networks (DNNs). On the Kedzierski dataset, MLStackXT achieved 95.85% accuracy, with a kappa score of 94.96%, F1-score of 95.73%, recall of 95.85%, and precision of 96.10%. Similarly, on the Jung dataset, the model attained 95.00% accuracy, a kappa score of 91.28%, an F1-score of 94.78%, a recall of 95.00%, and a precision of 94.81%. Confusion matrix analysis confirmed a significant reduction in misclassification, achieving 100% accuracy in 8 out of 12 MPs categories (Kedzierski) and over 97% accuracy in 4 out of 5 MPs categories (Jung). Model optimization via Optuna and interpretability analysis using SHAP highlighted the influence of principal component analysis (PCA) and key spectral features on classification performance. In addition, MLStackXT outperformed various stacking configurations, demonstrating its robustness and effectiveness for MPs detection.

Druh

J_imp - Článek v periodiku v databázi Web of Science

CEP obor

—

OECD FORD obor

20701 - Environmental and geological engineering, geotechnics

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

Journal of Environmental Chemical Engineering

ISSN

2213-2929

e-ISSN

2213-3437

Svazek periodika

13

Číslo periodika v rámci svazku

5

Stát vydavatele periodika

GB - Spojené království Velké Británie a Severního Irska

Počet stran výsledku

13

Strana od-do

nestránkováno

Kód UT WoS článku

001565151200021

EID výsledku v databázi Scopus

2-s2.0-105015139475