TracHack
Tracfone Wireless requested members of the UM MSBA program to develop a model that predicts one-time-redeemers
A one-time-redeemer is a customer that only uses one redemption than goes inactive
A machine learning algorithm was used in order to complete this task
Tracfone provided .json file with over 73,000 observations. Each observation was a customer ID number and variables about said customer. These independent variables ranged from SMS international usage to phone model to state purchased. There were 50 + different variables that could be included in the final model. We used Tracfone's servers to run models using different machine learning techniques. The results were calculated using a F1 Score. After weeks of testing different methods, variables, and a whole lot of debugging these are the scores that were achieved.
F1 Scores:
Linear Regression
.
0
Random Forest
.
0
Gradient Boosted Classification Tree
.
0