What is a retrain?
Retraining is a ‘business as usual’ practise for a predictive model. What a model retrain does is
- preserves the desirable features of the original model
- remedies identified shortcomings
- increases performance compared with original, and
- adapts the model to changes in input data since the first training.
With continuous changes to fleet makeup, ongoing reviews help identify emerging risk factors that will impact the model’s predictions, as well as provide an opportunity to enhance the model. The retrain allows us to pick up recent trends, and provides up-to-date information on the performance of all ships, class societies, flags, DOC companies and more.
Did we get the predictive model wrong the first time?
No – all predictive models constantly evolve and require revision to improve performance. Between each retrain there are incidents and events that affect how the model works. With continuous changes to fleet makeup, ongoing reviews help identify emerging risk factors that will impact the model’s predictions, as well as provide an opportunity to enhance the model. The retrain allows us to pick up recent trends, and provides up-to-date information on the performance of all ships, class societies, flags, DOC companies and more.
What changed in this retrain?
- a more recent data set has been used to train our algorithm and this has produced a different model to better respond to emerging trends and themes. A different tree leads to different predictions.
- the ship-type logic (which ships are included in the model) has been refined and no longer considers certain RORO / passenger vessel types and associated casualties. This has caused a recalculation of Regulator (Class & Flag), DOC & Builder variables and different outcome
- North Korean vessels have been removed from the model
- PSC logic has been updated. Previous models distinguished between those vessels with zero deficiencies at PSC (from inspection) and those vessels with zero deficiencies due to zero inspections. In responses to customer & Vetting Officer feedback that this is unfair, the logic has been changed to treat these as the same
What's in a good rating?
- Incident-free performance
- Strong PSC performance at vessel & DOC level
- Quality Flag State
- Quality Classification Society
- Quality DOC holder
- If there are problems – fix them!
If you have questions regarding a specific vessel please contact RightShip on email@example.com or call your local RightShip office.
Do current adjustments stay the same?
Yes. The current positive adjustment factors remain unchanged. They are:
- Certification– Green Award
- RightShip dry inspection
Did you get the retrain independently verified?
In addition to our own rigorous testing the retrain process has been independently reviewed by the renowned US-based Elder Research, who also reviewed and verified the original RightShip Qi predictive model.