What it does
This model predicts a t-mobile customer's satisfaction with their services and devices. The algorithm pulls data from a customer's billing history, length of period they have carried t-mobile, as well as their customer service interactions. The output of the algorithm can be used as an indicator for a variety of things, such as predicting the likelihood a user will go to a t-mobile retail store to if they may cancel their plan altogether.
How I built it
The algorithm was built using Microsoft Azure's Machine Learning and AI API, and is hosted entirely in the cloud. Data sets used to train it are open source cell coverage info from kaggle as well as proprietary t-mobile data surrounding customer service messages and customer history. The model is a single layer two class neural network trained by hyper-parameter tuning with an accuracy rating of ~92%.
What's next for Tmobile Customer Satisfaction Predictor
Determining the likeliest reason for a customer to be visiting a retail store on top of how likely they are to do such a thing would be the next step. After that adding sub models to analyze more aspects of the customer's history based on data attainable from their user account, as well as possible fraud detection would be logical directions for the algorithm to achieve a more holistic approach of a given user's behavior and feelings towards t-mobile.
Log in or sign up for Devpost to join the conversation.