Our notebooks have been initialized thanks to several sources provided around the Home Credit default risk Kaggle competition: data from Home Credit, Kernels-Notebooks from competitors, Discussions threads involving host and competitors. While the objective of competitors is to get the best AUC score Area under the ROC Curve by any means, we'll rather focuse on the ability to provide interpretable decisions, including the possibility given to customers to explore their case.
methodological notice (French) : P7_ML_for_Credit_Scoring.pdf slides : P7_Support.pdf
The Dashboard python file, can be found in the public repo : https://github.com/EtienneLardeur/Streamlit_App it is hosted remotely here : https://share.streamlit.io/etiennelardeur/streamlit_app/main/local_app.py
or download and rune from your folder : streamlit run local_app.py launched at localhost - working with remote inputs.
There are 5 different notebooks:
- P7_EDA: focusing on Exploratory Data Analysis,
- P7_FE: focusing on Feature engineering,
- P7_FS: the feature selection,
- P7_Model: focusing on scoring with model evaluation,
- P7_Interpretation : focusing on model interpretation ! also provide any attempt in order to properly design dashboard