Date of Completion

2023

Document Type

Honors College Thesis

Department

Plant Biology

Thesis Type

Honors College, College of Arts and Science Honors

First Advisor

Stephen Keller

Second Advisor

Nicholas Gotelli

Keywords

Centaurea, Machine Learning

Abstract

Abstract:

Proper taxonomic classification is important for biodiversity research and community ecology; however, it can be challenging for biologists to properly recognize and classify closely related species that visually appear very similar to one another. In the United States, the species Centaurea jacea ( C.jacea) and Centaurea nigra (C. nigra) and their hybrids are commonly found in New England, the Great Lakes and the Pacific Northwest. There is uncertainty regarding their identification and classification due to these species often interbreeding which results in hybrids that blur the species boundaries between Centaurea jacea and Centaurea nigra. Machine learning algorithms were utilized to address this problem and attempt to accurately predict species classification using direct measurement of a suite of putatively diagnostic capitula traits, images of capitula, and genetic ancestry. The algorithms used produced models with a wide range of accuracy that in their current working state could not be relied upon for a species classification. These algorithms are an emerging tool used for species classification in other study systems, but to our knowledge, no one has previously applied machine learning to classify Centaurea hybrid complexes.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS