Support Vector Machine Optimized by Genetic Algorithm for Data Analysis of Near-Infrared Spectroscopy Sensors

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Near-infrared (NIR) spectral sensors are crucial for material analysis, offering high precision in the tobacco industry.
• Spectral analysis using NIR sensors aids complex information processing and identification.

Purpose of the Study:

• To propose a novel method for discriminating tobacco cultivation regions using NIR sensor data.
• To enhance the discrimination capacity of support vector machine (SVM) models through effective input selection.

Main Methods:

• A support vector machine (SVM) model was developed for tobacco cultivation region discrimination.
• A genetic algorithm (GA) was employed for input subset selection to identify effective principal components (PCs) for the SVM model.
• Comparative experiments evaluated model performance using prediction accuracy and assessment criteria (true positive rate, true negative rate, positive predictive value, F1 score).

Main Results:

• The GA-SVM model demonstrated superior discrimination capacity compared to models using sequentially selected principal components (PCs).
• Certain PCs with less apparent information were found to be more effective for identifying cultivation regions.
• The proposed GA-SVM model effectively learned the relationship between tobacco cultivation regions and NIR sensor data.

Conclusions:

• The GA-SVM model offers an effective approach for classifying tobacco cultivation regions based on NIR spectral data.
• Input subset selection using GA is crucial for optimizing SVM model performance in this application.
• This method enhances the utility of NIR spectroscopy in agricultural applications, particularly for tobacco quality control and origin verification.