Evaluation Methodology

Evaluation Methodology: 

We invite you to submit your answer to the following tasks:

Task 1: 

Build a model that can predict which coordinates a vessel should prioritize in order to maximize the likelihood of catching the most of a specific type of fish. The prediction can be based on historical data and the size of the boat is not taken into account. 

Predict a point (coordinates) where most of this type of fish is likely to be caught on that day.

  • From submission day seven days into the future (26.09.2022-02.10.2022)
  • Ten different types of fish (Sild, Makrell, Hyse, Sei, Uer (vanlig), Lyr, Torsk, Breiflabb, Lange, Berggylt)
  • One possible location per fish per day (fangstdata files column Art FAO)

Submission format:

Csv file

Filename: teamname_sevendayspred.csv

Date;type of fish; lon; lat

09022022;Sild;55;5775

09022022;Torsk;55;5775

 

Metrics:

Correct or not correct prediction per fish per day in context of the maximum total fish weight in the area.
Precision, Recall, F1-score, MCC (MCC will be used for the ranking)
Total number of maximum possible caught fish type per predicted location (calculated by the organisers)

Task 2 and 3 will be evaluated by an expert jury consisting of industry, fishery and academic experts.


Task 2: 

Create a report of your analysis that can be read by experienced fishermen; an user-friendly visualization that a captain can read to make a assessment of where the vessel should search for fish the next day

Criteria that will be taken into account by the jury

  • Presentation
  • Usefulness
  • Correctness

Exploratory and innovative task with not right and wrong answers per se!


Task 3: 

Make a Sustainable Fishing Plan; a weekly plan that suggests the routes the fisherman should follow to optimize fish caught and fuel consumption 

Criteria that will be taken into account by the jury

  • Presentation
  • Usefulness
  • Correctness
  • Fish versus fuel consumption (subjectively assessed by fisherman)

Exploratory and innovative task with not right and wrong answers per se!

To compete for the prize awards all tasks are mandatory. Submission of only one sub-task is allowed, but will not be eligible for winning any of the prizes.

Click here to access an overview article over the FishAI Competition and datasets, published in the Nordic Machine Intelligence Journal

 

Submit your results via email to birte.hansen@nora.ai 

If you have any questions, please email Birte Hansen (birte.hansen@nora.ai)

Publisert 16. juni 2022 14:04 - Sist endret 20. sep. 2022 16:28