Enhanced data preprocessing with novel window function in Raman spectroscopy: Leveraging feature selection and...

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Data Science

Background:

• Accurate origin identification is crucial for agricultural product authentication.
• Raman spectroscopy offers a non-destructive method for chemical analysis.
• Traditional methods for origin identification can be labor-intensive and less accurate.

Purpose of the Study:

• To develop a robust and accurate method for raspberry origin identification.
• To explore the efficacy of combining novel spectral preprocessing, feature selection, and machine learning algorithms.
• To provide a reliable tool for agricultural product authentication and quality control.

Main Methods:

• Raman spectral data preprocessing using a window function combined with baseline removal.
• Optimization of the window function parameter (binning window width = 5).
• Application of feature selection techniques, with Information Gain showing superior performance.
• Construction and evaluation of predictive models using ten different machine learning algorithms.

Main Results:

• Optimized preprocessing significantly reduced data dimensionality and improved data quality.
• Information Gain effectively extracted discriminative spectral features.
• Linear Support Vector Classifier (LinearSVC), Multi-Layer Perceptron Classifier (MLPClassifier), and Linear Discriminant Analysis (LDA) achieved performance metrics above 0.96.
• Random Vector Functional Link Network Classifier (RVFLClassifier) achieved performance metrics above 0.93.

Conclusions:

• The proposed approach demonstrates high accuracy and robustness in identifying raspberry origins.
• The integration of advanced spectral preprocessing, feature selection, and machine learning is effective.
• This method provides a valuable tool for agricultural product authentication and quality control.