| Deliverable | Trained Model File |
| Machine Learning Task | Classification |
| Target Variable | Loan Status Default/Paid in Full |
| Control Metric | AUROC 0.73 and Confusion Matrix Examination |
This project showcases a risk analysis on loan payment status for the real world data from the Lending Club peer-to-peer lending platform. Lending Club make their data publically available.
This repository contains two notebooks that examine this data and analyse loan default prediction. This is a classification task with the aim of predicting which loans will default. The first notebook showcases some data exploration, viualization and feature engineering. The second notebook assess the performance of a set of standard classification algorithms against this dataset. Three classification algorithms were fitted against a datast of c. 80,000 loans. For risk-adverse investors/lenders the classification probability threshold was adjusted to reduce the numbers of False Negatives.
A detailed exploration of the results is provided including a demonstration of how to change the loan default probability thresholds for risk-adverse investors.
You can find the code for this project at: github repository