Bayesian Inference Modeling to Rank Cleanup System in Arctic oil spill
Situation
Selecting the correct response technology for emergency oil spill response is difficult. This work will rank response technologies (from MCR, CDU, & ISB) considering Arctic environement.
Task
- Since Oil Spill is rare in Arctic, we needed to build data pipeline to obtain reasonable numbers of incidents
- Build model to classify technologies
Action
- Based on Monte Carlo Simulation (implemented using distribution of feature variables, and outputs obtained from engineering model), 3100 scenarios is generated
- A #multi-class, multi-label classification system is developed. Bayesian Inference model is implemented using Naive Bayes Classifier. Multi-label: y = [y1, y2, y3] = [MCR, CDU, ISB]. Each label can have multiple classes e.g. [OK, Consider, Go Next Season, Unknown, Not recommended]
Result
Based on oil and environmental conditions in Arctic, our model proposes which technology would be better to respond oil spill. The model has 0.79, 0.93 and 0.93 ROC-AUC score for different technologies.
The picture below is an overview of the project's methodology. Further details can be found in this journal paper.