Kernel functions embedded in support vector machine learning models for rapid water pollution assessment via...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Machine Learning

Background:

• Water pollution poses significant environmental challenges.
• Near-infrared (NIR) spectroscopy offers rapid detection methods for water contaminants.
• Machine learning, particularly chemometrics, enhances the accuracy of NIR-based pollution assessment.

Purpose of the Study:

• To develop advanced calibration models for quantitative water pollution determination using NIR spectroscopy.
• To investigate the impact of different kernel functions within the Least Squares Support Vector Machine (LSSVM) algorithm.
• To propose and evaluate a novel logistic-based neural network kernel for improved LSSVM performance.

Main Methods:

• Utilized Least Squares Support Vector Machine (LSSVM) for establishing Near-infrared (NIR) calibration models.
• Investigated the influence of various kernel functions on LSSVM model accuracy for water pollution analysis.
• Developed and implemented a novel kernel function based on a logistic-based neural network, incorporating deep learning for parameter optimization.

Main Results:

• The novel logistic-based neural network kernel demonstrated improved prediction capabilities for assessing water pollution indicators like chemical oxygen demand.
• The proposed kernel enhanced model resistance to over-fitting, facilitating reliable cross-validation.
• The LSSVM method with the novel kernel proved effective for the quantitative determination of chemical oxygen demand.

Conclusions:

• A novel logistic-based neural network kernel offers a promising approach for enhancing LSSVM-based NIR spectroscopy in water pollution monitoring.
• This deep learning-integrated kernel provides a robust and accurate method for quantitative analysis of water quality parameters.
• The developed methodology holds potential for broader applications in water resource management and environmental monitoring.