Raman spectroscopy and machine learning for forensic document examination.

Area of Science:

• Forensic Science
• Analytical Chemistry
• Computer Science

Background:

• Document paper analysis is critical in forensic science for detecting forgery and fraud.
• Traditional methods for paper classification can be time-consuming and subjective.
• Developing objective and efficient methods for paper differentiation is essential.

Purpose of the Study:

• To classify document papers using Raman spectroscopy integrated with machine learning.
• To evaluate the performance of different machine learning models (Random Forest, SVM, FNN) for this classification task.
• To identify key spectral regions and preprocessing techniques that enhance classification accuracy.

Main Methods:

• Raman spectroscopy was employed to acquire spectral data from document papers.
• Machine learning models, including Random Forest (RF), Support Vector Machines (SVMs), and Feed-Forward Neural Networks (FNNs), were trained on spectral data.
• Spectral data underwent preprocessing, including first derivative transformation, and dimensionality reduction by focusing on the 200-1650 cm⁻¹ range.

Main Results:

• The Random Forest model demonstrated effectiveness in identifying important spectral features and regions.
• First derivative spectral preprocessing significantly improved classification performance across models.
• The Feed-Forward Neural Network model achieved the highest classification accuracy, with an F1 score of 0.968.
• The informative spectral range of 200-1650 cm⁻¹ reduced input variables while boosting model accuracy.

Conclusions:

• Combining Raman spectroscopy with machine learning provides an interpretable, efficient, and robust method for forensic document examination.
• The study highlights the potential of this integrated approach for accurate and reliable paper classification.
• This technique offers a promising advancement for forensic document analysis, aiding in fraud and forgery investigations.