Numerical taxonomy and genus-species identification of Czekanowskiales in China based on machine learning

Abstract

Czekanowskiales were the main component of the global Mesozoic flora and were sensitive to changes in the climate and environment during that period. However, accurate identification of Czekanowskiales fossils is difficult due to the similarities in some macroscopic and cuticular patterns among different genera and species. In the present study, a dataset of macroscopic and cuticular traits was collated based on the Czekanowskiales fossils from China. This study focused on the numerical taxonomy and identification of Czekanowskiales at the generic and species levels using cluster analysis, trait selection, and supervised learning methods for machine learning. Our results show that the studied 35 species can be clustered into three major groups, as consistent in a great extent with traditional taxonomic methods. Macroscopic traits are more important for the identification at the generic level, while cuticular traits are more valuable for the identification at the species level. The classification and regression tree as well as logistic regression algorithms demonstrated superior performance in the genus and species identification, and the inclusion of cuticular traits could significantly improve the accuracy of identification. This study provides quantitative analytical evidence for the taxonomy of Czekanowskiales fossils.