Session: Player Churn Prediction Using Machine Learning
Would you like to see the design and code behind a churn prediction machine learning solution?
Customer churn is a well-known term across many industries. Generally, it represents the act of a customer leaving a product for good. The goal of churn prediction is to identify such customers and try to prevent them from leaving by keeping them engaged in the product. This presentation will focus on churn prediction in mobile games where the goal is to identify the players that are the most likely to quit before they do. In most freemium games, only very few per cent of players spend their money on the game during their lifetime, which makes paying players even more valuable. Therefore, a successful churn prevention methodology is essential for a successful business in the freemium gaming industry.
In this talk, the churn prediction solution will be described together with the challenges we faced during their development. I will present how we approached data cleaning and preparation, what categories of features we created within feature engineering, which model type achieved the best results in prediction, and what metric we used to evaluate our models.
Bio: Nikola Valesova
Nikola Valesova is a full-time data scientist passionate about machine learning and fighting for diversity and inclusion. During her internships at Red Hat and Thermo Fisher Scientific, she gained valuable experience in programming in Python and C++ and image super-resolution using GANs. Currently, she works as a data scientist for a Prague-based start-up DataSentics, where her main responsibilities are the development of predictive models for the optimization of online advertising and churn prediction in mobile games.