awslabs/aws-customer-churn-pipeline

An End to End Customer Churn Prediction solution using AWS services.

Customer Churn Pipeline on AWS

A production-focused End to End churn prediction pipeline on AWS

It provides:

• One-click Training and Inference Pipelines for churn prediction
• Preprocessing, Validation, Hyperparameter tuning, and model Explainability all backed into the pipelines
• Amazon Athena and AWS Glue backend that allows for the pipeline to scale on demand and with new data
• End to End Implementation for your own custom churn pipeline

An AWS Professional Service open source initiative | aws-proserve-opensource@amazon.com

Table of contents

• Customer Churn Pipeline on AWS
  • Table of contents
  • Getting Started
  • Clean up
  • Read The Docs
  • Solution Architecture
  • Contributing

Getting Started

1. Step 1 - Modify default Parameters

Update the .env file in the main directory. To run with Cox proportional hazard modeling instead of binary logloss set COXPH to "positive".

S3_BUCKET_NAME="{YOU_BUCKET_NAME}"
REGION="{YOUR_REGION}"
STACK_NAME="{YOUR_STACK_NAME}"
COXPH="{negative|positive}"

2. Step 2 - Deploy the infrastructure

./standup.sh

3. Step 3 - Update the pending GitHub Connections

To configure the Github connection manually in the CodeDeploy console, go to Developer Tools -> settings -> connections. This is a one time approval. Install as App or choose existing.

4. Step 4 - Release change in churn pipeline for the first time

5. Step 5 - Once the build succeeds, navigate to Step Functions to verify completion

Note that as part of the build, it's going to run the Churn Training Pipeline.

6. Step 6- Trigger Inference pipeline. Batch Inference can be automated using cron jobs or S3 triggers as per business needs.

AWS_REGION=$(aws configure get region)
aws lambda --region ${AWS_REGION} invoke --function-name invokeInferStepFunction --payload '{ "": ""}' out

Clean up

./delete_resources.sh

This does not delete the S3 bucket. In order to delete the bucket and the contents in it, run the below -

source .env
 accountnum=$(aws sts get-caller-identity --query Account --output text)
 aws s3 rb s3://${S3_BUCKET_NAME}-${accountnum}-${REGION} --force

Read The Docs

Documentation

In addition, check out the blog posts:

• Deploying a Scalable End to End Customer Churn Prediction Solution with AWS!
• Retain Customers with Time to Event Modeling-Driven Intervention

Solution Architecture

Contributing

For how to Contribute see here.