This project is dedicated to the development of a machine learning model designed to estimate a customer's creditworthiness.
We begin with exploratory data analysis (EDA) of the credit_record.csv and application_record.csv datasets, which provide information about credit status and personal details of clients.
After a proper target estimation based on credit status information and data preprocessing, we continue with the Model Building phase, where multiple models are considered. We start with a Decison Tree model, gradually progressing to explore more complex models, including Random Forest, Balanced Random Forest, XGBoost and lightgbm. This comprehensive model selection process, which also includes the fine-tuning of the hyperparameters for each model, enables us to thoroughly assess their performance using a range of evaluation metrics, such as balanced accuracy and the F2 score, ensuring a meticulous examination of each model's predictive capabilities. The latter leads to the selection of the final model.
Finally, to gain a deeper understanding of the model's inner workings and to assess the significance of each feature in shaping the predictions for individual instances, we employ the SHAP technique, thereby shedding light on the hidden mechanisms at play within our machine learning model.
In summary, this project represents a comprehensive journey through the development of a creditworthiness estimation model, encompassing data exploration, model building, rigorous performance evaluation, and results' interpretability, all with the ultimate goal of providing valuable insights into the creditworthiness of our customers.
NOTE: I suggest to open the notebook with nbviewer at this link: https://nbviewer.org/github/CrisLap/Credit-worthiness-Estimation/blob/main/Creditworthiness%20project.ipynb