Overview / Usage

Our aim from the project is to make use of pandas, matplotlib, & seaborn libraries from python to extract insights from the data and xgboost, & scikit-learn libraries for machine learning.

Secondly, to learn how to hypertune the parameters using grid search cross validation for the xgboost machine learning model.

And in the end, to predict whether the loan applicant can replay the loan or not using voting ensembling techniques of combining the predictions from multiple machine learning algorithms.

Methodology / Approach

In this reference kit, we provide a reference solution for training and utilizing an AI model using XGBoost to predict the probability of a loan default from client characteristics and the type of loan obligation. We also demonstrate how to use incremental learning to update the trained model using brand new data. This can be used to correct for potential data drift over time as well to avoid re-training a model from full data via which may be a memory intensive process. Finally, we will provide a brief introduction to a few tools that can be used for an organization to analyze the fairness/bias that may be present in each of their trained models. These can be saved for audit purposes as well as to study and adjust the model for the sensitive decisions that this application must make.

Technologies Used

1. Python v3.9
2. XGBoost v0.81

Repository

https://github.com/DuelistRaj/loan-prediction

Collaborators