The Churn Preventer

Inspiration

Churn rate is a critical metric for companies whose customers pay on a recurring basis This metric is especially important for T-Mobile, which pays exceptional attention to customer experience

What it does

By predicting potential churning customers, T-Mobile would be able to look into those individuals closely and see what can be done to retain them

How I built it

Using the Telco Customer Churn dataset from Kaggle, we trained a neural network model with Scikit-Learn that predicts the likelihood of churning of customers based on selected features.

Challenges I ran into

Dealing with exceptional amount of data using neural network within 24 hrs is hard to tune the hyperparameters and adjust for the best result.

Accomplishments that I'm proud of

Nevertheless, we have split the grid search job across multiple devices, cross-compare the validation score, and constructed a web application, which is new to all of us.

What I learned

What I learned, again and again, is the importance of data, in the field of machine learning. Well defined and processed data is the first step to success.

What's next for The Churn Preventer

Periodic Boosting! With feedback from new customers (new data), we can use the technique of boosting to make our new generation model more powerful and precise.

Built With

• angular.js
• neural-network
• numpy
• python
• scikit-learn
• typescript