Customer Churn Prediction

Spotify: Predict telecom churn via multi-model benchmarking and drift-aware deployment. Python, Scikit-learn, XGBoost, Keras.

Assignment

You are provided with a sample dataset of a telecom company’s customers and it's expected to done the following tasks:

Perform exploratory analysis and extract insights from the dataset.
Split the dataset into train/test sets and explain your reasoning.
Build a predictive model to predict which customers are going to churn and discuss the reason why you choose a particular algorithm.
Establish metrics to evaluate model performance.
Discuss the potential issues with deploying the model into production.

The customer churn data is given in the file Data_Science_Challenge.csv. The detailed explanation is as follows:

Column Name	Column Type	Column Description
State	String	The state where a customer comes from
Account length	Integer	Number of days a customer has been using services
Area code	Integer	The area where a customer comes from
Phone number	Alphanumeric	The phone number of a customer
International plan	String	The status of customer international plan
Voicemail plan	String	The status of customer voicemail plan
No. vmail msgs	Integer	Number of voicemail message sent by a customer
Total day minutes	Float	Total call minutes spent by a customer during day time
Total day calls	Integer	Total number of calls made by a customer during day time
Total day charge	Float	Total amount charged to a customer during day time
Total eve minutes	Float	Total call minutes spent by a customer during evening time
Total eve calls	Integer	Total number of calls made by a customer during evening time
Total eve charge	Float	Total amount charged to a customer during evening time
Total night minutes	Float	Total call minutes spent by a customer during night time
Total night calls	Integer	Total number of calls made by a customer during night time

You’re free to use any programming language (Python, R, Julia) that you’re familiar with and include the code together with your analysis.