Geographical profiling of wood samples via ATR-FTIR spectroscopy and machine learning algorithms: Application in wood forensics

Illegal activities associated with deforestation for the lumber and furniture industries pose significant threats to plant and animal biodiversity, as well as natural resources. Accurate identification of wood sources is vital, yet traditional laboratory techniques often fall short in precisely determining the chemical composition of samples for classification. This study aims to leverage ATR-FTIR spectroscopy alongside machine learning algorithms to construct a robust model for discerning the geographical origins of wood samples from India. By systematically comparing various machine learning classifiers, we address the limitations of subjective visual interpretation and evaluate their accuracy using wood spectral data. Logistic regression emerges as the most effective classifier for distinguishing Eucalyptus (75 % accuracy), Dalbergia (68 % accuracy), and Populus (81.5 % accuracy) species. Through a methodology encompassing data pre-processing, classifier selection, and performance evaluation, this research offers promising tools for combating challenges posed by illegal wood trafficking and transportation. The outcomes hold significant potential for enhancing wildlife crime prevention efforts by facilitating the tracing illicit timber sources, apprehension of perpetrators, and implementation of preventive measures. •Geographical location identification of wood samples through quick and non-destructive ATR-FTIR spectroscopy.•Subjective biasness on analysis is reduced by applying machine learning algorithms.•PyCaret library in the Python framework is employed.•LR (Logistic Regression) Classifier outperforms among others.•External validation through unknown samples validates the best model.