kth.sePublications
3738394041424340 of 285
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
A Comparative Analysis of Unsupervised Learning Techniques for Identification and Prediction of Animal Behavior
KTH, School of Engineering Sciences (SCI), Mathematics (Dept.), Probability, Mathematical Physics and Statistics.
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
En jämförande analys av övervakade inlärningsmetoder för identifiering och förutsägelse av djurbeteende (Swedish)
Abstract [en]

While supervised machine learning has driven significant advancements in behavioral analysis, the capabilities of unsupervised learning methods remain largely unexplored. This thesis investigates the effectiveness of two such techniques in analyzing animal behavior: Variational Animal Motion Embedding (VAME) and Behavioral Segmentation of Open Field in DeepLabCut (B-SOID). Both methods are applied to a dataset generated by SciLifeLab, comprising recordings from three experimental animal groups, each paired with corresponding control groups. The detected behavioral motifs were subjected to comparative analysis to identify significant differences between treatment and control groups. Each method’s clustering quality was also evaluated using three different internal validity metrics: Silhouette Coefficient, Davies-Bouldin Index, and Cluster Density-Based Validity. The findings reveal that while B-SOID performs better on the validity metrics, VAME exhibits better cluster stability. Both methods identify similar behavioral differences within the data but excel in different contexts. B-SOID excels in rapid pose classification without capturing the temporal evolution of such a pose, whereas VAME outperforms in learning spatiotemporal information and embedding them in a dynamic latent space. These differences underscore the importance of selecting the appropriate method based on specific research objectives

Abstract [sv]

Medan övervakad maskininlärning har drivit på betydande framsteg inom beteendeana lys, förblir kapaciteten hos oövervakade inlärningsmetoder i stort sett outforskade. Detta examensarbete undersöker effektiviteten av två sådana tekniker för att analysera djurs beteende: Variational Animal Motion Embedding (VAME) och Behavioural Segmentation of Open Field in DeepLabCut (B-SOID). Båda metoderna tillämpades på en datauppsättning genererad av SciLifeLab, innefattande inspelningar från tre försöksdjursgrupper, var och en parad med motsvarande kontrollgrupper. De upptäckta beteendemotiven utsattes för jämförande analys för att identifiera signifikanta skillnader mellan behandlings- och kontrollgrupper. Varje metods klustringskvalitet utvärderades också med hjälp av tre olika interna validitetsmått: Silhouette Coefficient, Davies-Bouldin Index och Cluster Density-Based Validity. Resultaten visar att medan B-SOID presterar bättre på validitetsmåtten, uppvisar VAME bättre klusterstabilitet. Båda metoderna identifierar liknande beteendeskillnader inom datan, däremot utmärker dem sig i olika sammanhang. B-SOID utmärker sig i snabb positionsklassificering utan att fånga den temporala utvecklingen mellan poserna, medan VAME överträffar när det gäller att lära sig rumslig information och bädda in den i ett dynamiskt latent rum. Dessa skillnader understryker vikten av att välja lämplig metod utifrån specifika forskningsmål.

Place, publisher, year, edition, pages
2024.
Series
TRITA-SCI-GRU ; 2024:461
Keywords [en]
Unsupervised machine learning, Clustering, Behavioural analysis, VAME, B-SOID
Keywords [sv]
Oövervakad maskininlärning, Klustring, Beteendeanalys, VAME, B-SOID
National Category
Mathematical sciences
Identifiers
URN: urn:nbn:se:kth:diva-372318OAI: oai:DiVA.org:kth-372318DiVA, id: diva2:2011394
External cooperation
SciLifeLab
Subject / course
Mathematical Statistics
Educational program
Master of Science - Applied and Computational Mathematics
Supervisors
Examiners
Available from: 2025-11-04 Created: 2025-11-04 Last updated: 2025-11-04Bibliographically approved

Open Access in DiVA

fulltext(4048 kB)5 downloads
File information
File name FULLTEXT01.pdfFile size 4048 kBChecksum SHA-512
441f0087dc8c19e4cf4de9a9a72a944a2ec51479d39af40973afef2408a27c6583093a135a130ada054eb4cc705283646767502e07deb0106be041ff02f45b33
Type fulltextMimetype application/pdf

By organisation
Probability, Mathematical Physics and Statistics
Mathematical sciences

Search outside of DiVA

GoogleGoogle Scholar
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 123 hits
3738394041424340 of 285
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf