High-level data fusion using majority voting for the classification of spray paint spectroscopic data.

Area of Science:

• Forensic Science
• Analytical Chemistry
• Machine Learning

Background:

• Forensic paint analysis relies on distinguishing between paint samples.
• Multimodal spectral data offer complementary information for paint characterization.

Purpose of the Study:

• To investigate a high-level data fusion approach for classifying spray paint samples.
• To evaluate the combined discriminating ability of multiple analytical techniques and machine learning classifiers.

Main Methods:

• Collected spectral data using Fourier transform infrared (FTIR) spectroscopy, Raman spectroscopy, scanning electron microscopy coupled with energy-dispersive spectroscopy (SEM-EDS), and UV-Vis microspectrophotometry (MSP).
• Modeled data using Naïve Bayes, k-nearest neighbors (KNN), support vector machine (SVM), random forests, and extreme gradient boosting (XGBoost) classifiers.
• Integrated intermediate predictions using majority voting for a high-level data fusion scheme.

Main Results:

• The data fusion approach consistently outperformed individual instruments in classification accuracy.
• Near-perfect classification was achieved, especially for red and blue paints.
• Random Forest and Naïve Bayes classifiers demonstrated the most stable performance.

Conclusions:

• Fusing complementary spectral information enhances discriminative ability and reduces redundancy in paint analysis.
• The developed framework is computationally efficient, reproducible, and mirrors forensic examiners' processes.
• This approach offers a promising pathway for integrating multimodal spectral data in forensic paint examinations.