At GDC 2015, Ninja Metrics CEO Dmitri Williams gave a lecture that addressed a series of business issues facing marketers, BI teams and developers in the gaming industry. This lecture, Analytics 201, builds upon the intro material to Analytics 101: Definitions, and covers topics including basic knowledge of Machine Learning Models (MLMs), pros and cons of regression, Lifetime Value (LTV) modeling, Customer Acquisition Cost(CaC), Network models, Attribution approaches and empirical benchmarks.
Regression vs. Machine Learning
Regression is one of the most commonly used models for examining the relations among several variables. Although it is very straightforward and has a simple equation, it can be misinterpreted since it often times includes many general variables, and reflects correlations instead of the cause and effect. Therefore, when you find your regression lack of explanatory power, you may want to turn to machine learning to get more accurate results.
Machine learning is the process in which you leverage a program to analyze data and discover patterns in order to predict future events. Three commonly used machine learning models are Rule Set (JRIP, FOIL, others), Decision Tree Model and Support Vector Machine.
Machine learning can be very accurate, but the result might not be understandable and actionable, because what you usually see in the results are just several lines of abstract data or rules without context. Therefore, it’s important to have one domain expert who clearly knows the background and goals of the project on the team, to make sure that all variables put into analysis are selected based on quantitative or qualitative methods.
To learn about other behavior measurements such as LTV and Churn Modeling, LTV vs. CaC, and Network Models, you can watch Dmitri’s Game Analytics 201 speech at GDC 2015 here.