Hello all this is a public repo that has the approach for JOBATHON_JAN23 with Private leaderboard rank of 10 Approach:
- Data Preprocessing and Feature Engineering:
- Add more features using feature concatenation on all Categorical data.
- Feature Interactions between claim amount and vintage like divide and multiply were used.
- Binning the numerical columns like Claim amount.
- Using Mean encoding of Target by leaking it to the test data.
- Label Encoding
- Feature Selection:
- Sequential Feature selection using Xgboost
- Tuning:
- Tuning both Xgboost and Catboost models with best features.
- Blending:
- Final Blending of XGBOOST and CATBOOST with 0.8, 0.2 as co-efficients.