FishAI: Predicting high-likelihood Atlantic cod locations utilizing an ensemble Lasso regression model on geospatial data

Bugel, Lieke (2024) FishAI: Predicting high-likelihood Atlantic cod locations utilizing an ensemble Lasso regression model on geospatial data. Bachelor's Thesis, Artificial Intelligence.

Preview	Text FishAIthesiss4341724final.pdf Download (1MB) | Preview
	Text Toestemming.pdf Restricted to Registered users only Download (235kB)

Abstract

This study evaluates the performance of an ensemble Lasso regression model in predicting high-likelihood Atlantic cod locations and corresponding catch yield within the Norwegian Exclusive Economic Zone using historical catch data and environmental data. An ensemble Lasso regression model, consisting of multiple meta-models, was selected for this study, with its architecture providing both temporal and spatial context to the predictions. A sliding window technique generates input-output datasets, enabling supervised learning methods to capture temporal dependencies from the data. Each window consists of 5 input days of historical catch and environmental data, along with 5 output days of cod catch locations and associated product weights, with the model learning from these input-output pairs for future predictions. While the model successfully identified general trends in daily cod distributions, it showed signs of underfitting, as indicated by the simplified S-shaped predicted patterns and low R² scores. The mean squared error values indicate relatively low prediction error, but denormalized location and weight deviations highlight uncertainties about the model's practical utility. A critical limitation was the lack of spatial and temporal context in the input data.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Cardenas Cartagena, J. D.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	18 Dec 2024 10:43
Last Modified:	18 Dec 2024 10:43
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/34510

Actions (login required)

View Item