Estimating Customer Lifetime Value in the Gaming Industry Using Incomplete Data

Abstract

We were asked by Innovation Embassy to work with a large dataset centred around gambling investment, with the task of making a predictive function for computing Customer Lifetime Value (CLV), and also to see if there are ways of detecting fraudulent financial practices and addictive gambling patterns. We had moderate success with the data as it stands, but we were partly held back for two main reasons: the ability to discern a solid definition of CLV due to highly inconsistent data and data that contained many large and incomputable gaps. Different machine learning algorithms were used to find CLV functions based on key variables. We also describe a short and explicit list of ways where the base data can be improved to support effective calculation of CLV. Our key findings suggest that the average customer's CLV is 1035 and ~80% of revenue is brought in from ~10% of the clients.

A stacked ensemble learning method for customer lifetime value prediction

“…Module K-Nearest Neighbor (KNN) imputer with 2 neighbors was used to fill in missing values to solve this problem. KNN Imputer produces a more precise estimation of missing values because it fills the missing values by considering the average of corresponding data located in the nearest neighbors (Ali et al. , 2021).…”