acceleratoRs/CreditRiskPrediction

Data Science Accelerator - Credit Risk Prediction

The intention of this AcceleratoR is to provide material as a kickstart for a data scientist initiating a project in credit risk prediction, first using R and then extending this with using Microsoft R for any size datasets. The supplied smalle example random dataset is used to illustrate the process. It can not be used to provide actual financial insights as it is a randomly generated dataset. A data scientist can first replicate the process using the dataset supplied here and then replace the datasets with their own actual datasets and replicate the processing, tuning it to suit their own needs, as the starting point for the advanced development of machine learning models for credit risk prediction.

Introduction

Credit Risk Scoring is a classic but increasingly important operation in banking as banks continue to carefully monitor risk when lending for mortgages, credit cards, or commercial purposes. The industry remains very competitive. Accurate credit risk scoring enables a bank to predict the likelihood of default on a transaction. This will in turn help evaluate the potential risk posed by lending money to consumers and to mitigate losses due to bad debt as well as determine who qualifies for a loan, at what interest rate, what credit limits, and even to determine which customers may be interested in extended products.

Many banks nowadays are driving innovation to enhance risk management. Typically banks are exploring opportunities to segment millions of active credit card customers to improve risk scoring to then identify opportunities to offer increased limits, for example. Using advanced analytics for credit risk scoring exends traditional scorecard building and modelling with machine learning ensembles. Beyond this they are also pursuing innovations with customer oriented aggregation of transactions, multi-dimensional customer segmentation, and conceptual clustering to identify multiple segments across which to understand bank customers.

This repository contains three folders:

• Data This contains random artificial sample data with which we can demonstrate the process.

• Code This contains the R code and modelling process using R markdown files which can be converted to various formats, including pdf, html, ipynb, etc.

• Docs This contains miscellaneous documents including a blog post and installation instructions.

Business domain

Finance and risk analysis, credit risk prediction.

Data science problem

Normally credit risk prediction is categorized as a classification problem: given the customer transaction records and their demographic data, the task is to predict whether the consumer will default on a transaction in the near future, what is the default likelihood, and which factors are most likely to cause the default. Given this knowledge business can make decisions to improve their performance.

Data understanding

In the data-driven credit risk prediction model two types of data are normally taken into consideration.

1. Transaction data The transaction records cover transaction_id, account id, transaction date, transaction amount, merchant industry, etc. This transaction-level data is usually aggregated to provide transaction statistics and financial behavior information at the customer account level.

2. Demographic and bank account information This type of data captures the characteristics of the individual customer or account credit bureau, such as age, sex, income, and credit limit. They are generally static or change little over time.

Feature Engineering

1. Customer-oriented transaction aggregation is conducted to generate features which capture transaction dimensions at industry level and emphasize customer financial behaviors.

2. Binning analytics are optionally chosen to recode variables, thus eliminating the effect of extreme values.

Modeling

1. Traditional logistic regression model with L1 regularization are built as a baseline.

2. Machine learning models, such as gradient boosting and random forest, or their meta-ensembles, are fine tuned to compare the performance and choose an apporiate model to deploy.

3. An innovative AI approach using a evolutionary hotspot method is also being pursued and will be incorporated in a future release. This method builds multiple segmentations to build unique profiles for each customer from which we can reason about the customer's behaviours.

Scalability

Faster and scalable credit risk models are built using the state-of-the-art machine learning algorithms provided by the MicrosoftML package.

Operationalization

An R model based web service for credit risk prediction is published and consumed by using the mrsdeploy package that ships with Microsoft R Client and R Server 9.1.

Application Development

A Credit Risk Application through a REST API is developed by integrating the published web service with a shiny framework.