Owoeye, Kehinde; (2021) On computational models of animal movement behaviour. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text transfer_report.pdf - Accepted Version Download (28MB) | Preview

Abstract

Finding structures and patterns in animal movement data is essential towards understanding a variety of behavioural phenomena, as well as shedding light into the relationships between animals among conspecifics and across different taxa with respect to their environments. The recent advances in the field of computational intelligence coupled with the proliferation of low-cost telemetry devices have made the gathering and analyses of behavioural data of animals in their natural habitat and in a wide range of context possible with aid of devices such as Global Positioning System (GPS). The sensory input that animals receive from their environment, and the corresponding motor output, as well as the neural basis of this relationship most especially as it affects movement, encode a lot of information regarding the welfare and survival of these animals and other organisms in nature's ecosystem. This has huge implications in the area of biodiversity monitoring, global health and understanding disease progression. Encoding, decoding and quantifying these functional relationships however can be challenging, boring and labour intensive. Artificial intelligence holds promise in solving some of these problems and even stand to benefit as understanding natural intelligence for instance can aid in the advancement of artificial intelligence. In this thesis, I investigate and propose several computational methods leveraging information theoretic metrics and also modern machine learning methods including supervised, unsupervised and a novel combination of both towards understanding, predicting, forecasting and quantifying a variety of animal movement phenomena at different time scales across different taxa and species. Most importantly the models proposed in this thesis tackle important problems bordering on human and animal welfare as well as their intersection. Crucially, I investigate several information theoretic metrics towards mining animal movement data, after which I propose machine learning and statistical techniques for automatically quantifying abnormal movement behaviour in sheep with Batten disease using unsupervised methods. In addition, I propose a predictive model capable of forecasting migration patterns in Turkey vulture as well as their stop-over decisions using bidirectional recurrent neural networks. And finally, I propose a model of sheep movement behaviour in a flock leveraging insights in cognitive neuroscience with modern deep learning models. Overall, the models of animal movement behaviour developed in this thesis are useful to a wide range of scientists in the field of neuroscience, ethology, veterinary science, conservation and public health. Although these models have been designed for understanding and predicting animal movement behaviour, in a lot of cases they scale easily into other domains such as human behaviour modelling with little modifications. I highlight the importance of continuous research in developing computational models of animal movement behaviour towards improving our understanding of nature in relation to the interaction between animals and their environments.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item