Worked with security services team on a binary classification XGBoost model to predict whether a trade would settle within the contractual settlement date and improved the model performance from 45% to 55% F1-score and reduced the computational time by around a millisecond per trade using exploratory data analysis and feature engineering.

• Used Google Maps API to find the user’s location and made location based features to add those features in the existing boosted model to predict user's days past due to increase the model performance from 73% to 81% F1- score.• Clustered regions based on default rate in last one year and build a model that gives an alert if an application comes from a risky area which helped the company to prevent 16 probable fraud attacks made from July 2019 to June 2020.