santander-pr's Introduction

Santander Product Recommendation

File Structure

Santander-PR/
├── Models
│   ├── #1-Voting
│   │   ├── reproduce
│   │   │   ├── Product-Wise
│   │   │   │   ├── product-wise-lag.ipynb
│   │   │   │   ├── product-wise-lag.py
│   │   │   │   └── proposed-sampling-month.pptx
│   │   │   └── XGBoost
│   │   │       ├── xgboost.ipynb
│   │   │       └── xgboost.py
│   │   ├── voting.ipynb
│   │   └── voting.py
│   └── EDA
│       ├── EDA.ipynb
│       └── EDA.py
├── README.txt
└── Santander_Product_Recommendation.pdf

The report is saved as 'Santander_Product_Recommendation.pdf'
We have submitted the best model. You can find them in 'Models' folder
The top submission of ours is in 'Voting folder'
For Voting we had used multiple models to get a consolidated score.
To reproduce those CSVs please look at 'reproduce' folder
In 'reproduce' folder you can find two sub folders XGBoost and Product-Wise
'Product-Wise' folder needs to be reproduced for every product separately.
- We had used the Product Wise analysis to check which month is most suitable for training and trained accordingly
- But as mentioned in the report, the accuracy is more dependant on the latest month than this.
- We have also attached a PPT consisting of the inferences we found from running all such notebooks
On running the 'XGBoost' code we get it's respective CSV
Each folder has it's jupyter notebook(for better visualisation) and it's relavent python code.
Incase you want to skip the part where you run all the notebooks to get the CSVs, we have CSVs stored in drive