Time Series for Cheat Detection?
As to my personal knowledge about Minecraft ACS (see
ODIN - Next Generation Anti Cheat for an overview on ACS in the past), time series data has never been considered a viable option when it comes to cheat detection. The aim for ODIN is to change this, since time series data can in fact effectively be used to classify cheaters, as will become apparent in this blog post. Before we begin, I’d like to give a small overview of what we will be doing in this development log.
- We start with an explanation on time series and how they can be used for cheat detection.
- Next, a simplified time series implementation will be presented that is a direct copy of the one being used in ODIN.
- Once a time series is defined, we will discuss the different possible routes towards using them in an ACS.
- Finally, some examples on the classification of time series will be provided to give an insight in how ODIN works internally.
The goal is not to show the full internal workings of ODIN (since this would require me to write a complete book). Rather, I want to give the people who are reading this (you) an idea on what next-generation steps in ACS might be. In this particular case, this will be achieved by looking at time series and some simple models that can be used. In practice, ODIN implements many more models and aggregates the results of these models to construct prediction intervals (this will be discussed in a later development log) used for player classification.
Before we can jump to the interesting details on how time series work in ODIN and how they are used to classify players as cheaters, we first have to construct a solid foundation of time series. A time series will from now on be considered as a sequence of data points which are measured over successive equal intervals. To keep this development log understandable, we will not dive into the deep mechanics of what a time series represents mathematically.
Characteristics Of Time Series
Before showing a practical example involving time series data, it will first be useful to broaden the context in which the analysis of such data applies. In the case of ODIN, time series analysis could prove to be a useful tool, especially in the detection of movement- and combat-like cheats. In this development log, we will put the focus on movement, since it allows for the use of intuitive features such as movement, speed, etc.