predictCustomerChurn

Predict Customer Churn

This demo involves predicting customer churn for a telecommunications company.

Highlights

Model Creation

• Using Jupyter Notebooks to create and score a model
• Use DSX specific features like asset sharing, collaboration

Model Deployment

• Save and retrieve models using Cloud Object Storage
• Deploy the final model in a Watson Machine Learning repository

Application Deployment

• Access deployed model through an API call to WML
• Deploy a web app on Bluemix that uses the accessed model

Project-Specific Information

Tools Used

• Python, through Jupyter Notebooks
• Spark, to allow for scaleable modelling through SparkML
• Data Visualization through IBM Pixiedust

Model Information

• Random Forests Classifier, trained using SparkML pipelines
• Response variable is Customer Churn
• Features include age, income, gender, mobile data usage, payment method, etc.

Relevant Files

Notebooks

• Predict Customer Churn.ipynb - trains the initial model
• Predict Churn - Score New Data.ipynb - performs testing on unseen data
• Predict Churn - Deploy to WML.ipynb - interfaces with Watson Machine Learning to deploy the trained and scored model

Data Assets

• churn.csv - churn, by customer ID
• customer.csv - customer information, including ID
• customer_churn.csv - a combined data table with customer information and churn
• new_customer_churn_data.csv - unseen data, with only customer information, but no churn value