Client wanted to test and build an innovation proof of concept to predicting the valid hit (Rule-in hit), using the historical patterns of the travellers and profile available in the system (idea was to improve accuracy of predictions in comparison to rule based approach that is deployed currently).
A machine leaning based approach was adopted to reduce the False Positive rate for the hit generation. The objective was to increase potential rule-in hit cases of new travellers crossing the border.
- Leveraged open source R, Python for modelling
- Clustering, Time Series & Regression algorithms used
- Increased probability of identifying a risky passenger
- False Positive Rate is improved by 77%
- Overall hit accuracy is also improved by 2 bps