Classificação de argila utilizando espectroscopia na região do infravermelho médio e algoritmos de aprendizagem de máquina

Abstract

The characterization and classification of natural ceramic materials, such as clay, are essential for archaeological, environmental, and industrial applications, requiring fast and accurate methods. This work aimed to classify clay samples collected near the Alto do Bonfim archaeological site (GO) using Fourier Transform Infrared (FTIR) spectroscopy combined with multivariate analysis and machine learning algorithms. Three samples (A1, A2, and A3) were analyzed by FTIR, and the resulting spectra were preprocessed using Standard Normal Variate (SNV), followed by dimensionality reduction through Principal Component Analysis (PCA) and classification using K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA) algorithms. Despite the visual similarity of the spectral data, the models gradually revealed a separation - especially for sample A3, which has a distinct chemical composition. The analysis indicated that the 600–1500 cm⁻¹ spectral region contains the most relevant variability for discrimination. The LDA algorithm achieved 100% accuracy in both spectral ranges, outperforming KNN and SVM, and proving to be the most effective method. It is concluded that the combination of FTIR, PCA, and LDA provides a robust and efficient approach for automated clay classification, with potential applications in various scientific fields.