Team Memebers: Hongdou Li, Jacques Sham, Katja Wittfoth
As a part of a final project for Advances Machine Learning class at USF we were given a data set from the company LeanPlum with a goal to predict user churn. It is usually too late for a company when a user is churned. Therefore, the task of this project was to recognize the early signs of churn by predicting to predict whether or not those users would make a purchase within the next 7 or 14 days. The prediction is a binary classification.
The data describes attributes, session behavior, and records of events of 600,000 unique users. After we have performed an EDA on the given data, we extracted various features and combined them with labels.
We applied various Machine Learning algorithms to predict users’ purchases. We built several models including Logistic Regression, RandomForest, Gradient Boosting, and Neural Networks. XGB Boost Classifier showed the best result. After tuning the model, we improved the initial algorithm performance and were able to achieve AUC of 0.986.
