Can quantitative wood anatomy data coupled with machine learning analysis discriminate CITES species from their look-alikes?

• Autor(a) principal: Liu, Shoujia
• Data de Publicação: 2022
• Outros Autores: He, Tuo, Wang, Jiajun, Chen, Jiabao, Guo, Juan, Jiang, Xiaomei, Wiedenhoeft, Alex C. [UNESP], Yin, Yafang
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1007/s00226-022-01404-y; http://hdl.handle.net/11449/240628

Due to increasing global trade of timber commodities and illegal logging activities, wood species listed in the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) appendices are facing extinction, and their international trade has been banned or is under supervision. Reliable and applicable species-level discrimination methods have become urgent to protect global forest resources and promote the legal trade of timbers. This study aims to discriminate CITES-listed species from their look-alikes in international trade using quantitative wood anatomy (QWA) data coupled with machine learning (ML) analysis. Herein, the QWA data of 14 CITES-listed species and 15 of their look-alike species were collected from microscope slide collection, and four ML classifiers, J48, Multinomial Naïve Bayes, Random Forest, and SMO, were used to analyze the QWA data. The results indicated that ML classifiers exhibited better performance than traditional wood identification methods. Specifically, Multinomial Naïve Bayes outperformed other classifiers, and successfully discriminated CITES-listed Pterocarpus species from their look-alike species with an accuracy of 95.83%. Furthermore, the discrimination accuracy was affected by the combinations of wood anatomical features, and combinations with fewer features included could result in higher accuracy at the species level. In conclusion, the QWA data coupled with ML analysis could unlock the potential of wood anatomy to discriminate CITES species from their look-alikes for forensic applications.