dc.description.abstract | This study investigates the potential of using Automatic Identification System
(AIS) data to predict fish processing at Grieg Seafood Stjernelaks. The re-
search involved creating and analyzing two processed datasets: Labeled Days
and Labeled Time Series. The Labeled Days dataset uses the Active label,
indicating the days when fish processing occurred, while the Labeled Time Se-
ries dataset uses the Direct label, indicating the specific times when fish was
directly delivered by relevant vessels. Machine learning techniques, including
feature engineering, decision trees, random forests, and dynamic time warping,
were used to analyze the AIS data.
The results of this study highlight that the Labeled Days baseline utilizing
temporal patterns to predict the activity status for Stjernelaks perform excellent
in terms of Area Under the ROC Curve (AUC-ROC ) score. However, the best
machine learning model, ’Rand RFE RF,’ outperforms the baseline by utilizing
AIS data with an AUC-ROC score of 0.933. No model outperformed the
baseline for the Labeled Time Series dataset.
The study concludes that while AIS data shows promise in predicting if Stjer-
nelaks is processing fish on any given day, it does not conclusively prove that
AIS can be used to predict fish processing at Grieg Seafood Stjernelaks. The
research faced limitations due to issues encountered with Kystdatahuset’s API
endpoint for fetching AIS data, and the scarcity of label data. These limitations
may have affected the ability to fully answer the research question and should
be addressed in future research.
ii | |