Explainable AI for Financial Advisory: Churn Prediction

This repository hosts a prototype of an Explainable AI platform for Financial Advisory. The platform aims to provide transparent insights for customers who are at risk of leaving a financial service. The prototype includes three churn prediction models (Logistic Regression, Random Forest and KNN) along with explanations generated using SHAP.

Repository Contents

data/
├─ AdvisoryPerformance.csv
├─ Clients.csv
├─ clients_dataset.pickle
├─ Contracts.csv
├─ MarketPerformance.csv
├─ Transactions.csv
.gitignore
churn_prediction_models.ipynb
prepare_clients_dataset.py
requirements.txt

Please note that .csv files are not included due to GitHub file size limits.

The churn_prediction_models.ipynb notebook conducts evaluation and explanation of the churn prediction models. The required dataset to run the notebook, clients_dataset.pickle, is already included in the data/ directory. If you wish to generate the dataset from the original data, download the original .csv files, place them in the data/ directory as shown in the repository contents tree above, and execute the prepare_clients_dataset.py script.

Requirements

To execute the churn_prediction_models.ipynb notebook or the prepare_clients_dataset.py script, you will need a Python 3.8.10 virtual environment. Follow the instructions below to set up the virtual environment using a Windows Command shell:

1. Create a virtual environment named .venv: python -m venv .venv
2. Activate the virtual environment: .venv\scripts\activate.bat
3. Install the required Python packages: pip install -r requirements.txt
4. Using the .venv virtual environment, run the script from the Windows Command shell or execute the notebook using your preferred notebook interface.