keras-customer-churn

This repository contains a Shiny application that demonstrates the use of the customer churn model described in the Using Deep Learning With Keras To Predict Customer Churn blog post.

Maintenance Status

This repository is kept as a historical example of a Shiny application backed by a Keras customer churn model. It has not been actively maintained since 2019, so dependencies and deployment instructions may need updates before production use.

Using the Application

The application is deployed online at https://jjallaire.shinyapps.io/keras-customer-churn/.

To run the application locally, clone the repository then:

rmarkdown::run("customer_churn.Rmd")

The shiny application has three tabs:

1. Customer Scorecard---Analyzes a single customer at a time. The keras model is used to return the probability of customer churn. The app then recommends three strategies to mitigate churn risk:

   • Main Strategy - Incorporates tenure, contract type, key services, monthly charges to recommend offerings that reduce churn risk
   • Commercial Strategy - Incorporates specific services that the customer may be interested in
   • Financial Strategy - Incoporates payment method recommendations to reduce churn

2. Churn Facets---Analyzes aggregate churn by various features including type of contract, revenue, tenure and internet service. Drop-box filters are available to subset the data and drill into important customer segments.

3. Correlation Analysis---The correlation analysis shows the features that correlate to churn, which is important for a global perspective of understanding what affects churn.

Training the Model

Use the customer_churn.R script to train the model used by the application from scratch:

source("customer_churn.R")