To access this work you must either be on the Smith College campus OR have valid Smith login credentials.
On Campus users: To access this work if you are on campus please Select the Download button.
Off Campus users: To access this work from off campus, please select the Off-Campus button and enter your Smith username and password when prompted.
Non-Smith users: You may request this item through Interlibrary Loan at your own library.
Publication Date
2021
First Advisor
Annaliese Beery
Document Type
Honors Project
Degree Name
Bachelor of Arts
Department
Biological Sciences
Keywords
Machine learning, Social behavior, Voles, DeepLabCut, Pose analysis software
Abstract
One of the most important tests of selective social relationships in laboratory animals is the partner preference test (PPT). This test quantifies the preference of a certain animal for a familiar individual by comparing the time it spends with that animal to the time it spends with a stranger over a three-hour period. Though effective and useful, the PPT is time consuming and labor intensive to score by hand. In this project, we attempted to use free deep learning programs to create a process to automatically score partner preference tests. We used DeepLabCut (pose analysis software) and SimBA (behavioral analysis software) to track the movements of voles (Microtus sp.) as they moved around a preference test arena. We compared four different test models with various specifications to determine which was the most accurate in identifying individual animals and in correctly identifying body parts. We found that the programs were uniformly mediocre at identifying body parts, and the models made in multi-animal DeepLabCut were significantly worse than the single-animal DeepLabCut model at identifying individual animals. As the identification of individual animals is a crucial step in creating a program that can score a PPT, we were unfortunately unable to create a working version. However, when new releases of DeepLabCut and SimBA improve individual identification, the foundational work of this project can be used to create a process to make it easier and more efficient to analyze partner preference tests and so contribute to the future work of the lab.
Rights
©2021 Annie Aurora Dobroth
Language
English
Recommended Citation
Dobroth, Annie Aurora, "Machine learning approaches to multi-animal behavioral analysis" (2021). Honors Project, Smith College, Northampton, MA.
https://scholarworks.smith.edu/theses/2333
Smith Only:
Off Campus Download
Comments
39 pages : illustrations (chiefly color) Includes bibliographical references