Supervised Machine-Learning as a Decision Support Aid in Sea Lice Control for Norwegian Salmon Farmers

Streckfuss, Julia (2023) Supervised Machine-Learning as a Decision Support Aid in Sea Lice Control for Norwegian Salmon Farmers. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (1MB) | Preview

Abstract

In the country’s goal of five-fold salmon production expansion until 2050, the biggest salmon producing nation, Norway, has put the strictest average female lice thresholds globally in place to adhere to standards of sustainable aquaculture. Previously conducted research has had the focus to simulate lice dispersal especially with the aim of estimating lice pressure for passing wild salmon fish, however no available academic research, to the knowledge of the researcher, exists which helps salmon farmers to comply with government-set lice thresholds. The presented study f ills this gap by utilizing supervised classification models to identify the best point of warning to take preventative counter-actions for lice threshold exceedance, to classify whether farm localities are expected to exceed thresholds for future points in time, and to propose treatments proven most successful to them as well as to provide risk reduction estimates to avert the risk of exceeding if this is achievable. The focus of the study therefore lies at predicting the risk of exceedance of government-set lice thresholds at individual farm level and to provide treatment recommendations at a time when exceedance can still be prevented, which, to the knowledge of the researcher, has not been attempted by any other study published in the research domain yet. Furthermore, this study is the first study, to the knowledge of the researcher, which uses machine learning, in particular classification models, for the purpose of estimating lice counts at individual farm level in Norway and beyond. The strongest performing models have been found to be Random Forest, XGBoost, and AdaBoost with ROC-AUC scores between 0.997 and 0.985. Little to no degradation of models was found in comparing classifier performance from warning point 4 to 8 weeks prior to exceedance, and while metrics are similarly strong for both precision and recall, the preferred methods show slightly higher scores of precision than recall which demonstrates their ability to keep false positives low to mitigate the risk of unnecessary treatment costs encouraged.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Mulwa, Catherine UNSPECIFIED
Subjects:	G Geography. Anthropology. Recreation > GC Oceanography Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science H Social Sciences > HD Industries. Land use. Labor > Specific Industries > Agriculture Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Tamara Malone
Date Deposited:	06 Jan 2025 17:54
Last Modified:	06 Jan 2025 17:54
URI:	https://norma.ncirl.ie/id/eprint/7277

Actions (login required)

View Item