The overall flow of the analysis is summerized with the help of a flow chart
- /model : XGBoost saved model
- /data :
- raw_data/ : original dataset
- interim_data/ : processed dataset with added variable 'target'
- /doc :
- report : a quick report that summarizes the whole analysis
As, we find our data is highly imbalanced, we focussed on ROC AUC metrics for each model. Individual models metrics are
| Model | Test data AUC |
|---|---|
| Random Forest | 0.908558 |
| XGBoost | 0.964785 |
As, data is highly imbalance, our area of focus is high Recall. We build a final contingency table for each model to analyse the buckting of the ranks.
| Random Forest | XGBoost |
|---|---|
 |
 |
By compairing the contigency table we can see XGBoost performs well in bucketing majority to the lower rank of 1.
- Feature Enginerring
- Feature Selection
Can put more light on Feature Engineering by creating more important features and Feature Selection to find the important subset of features related to different models.