In this talk, we will explore and learn how to use the package Breakout Detection, developed by some folks at Twitter. This package detects level shifts in univariate (often time series) data. We will go through some of the math behind the algorithm, and then look at some examples of this algorithm in action.
In this talk we will explore how to use the Breakout Detection to detect level shifts in univariate data. Roughly, a level shift is a significant, sustained change in the mean of the data. Level shifts can be a signal that some fundamental change in environment the data is living in has occurred. For instance, the number of Tweets about the show Breaking Bad as a function of time as new episodes came out, or as evidenced in price of the S&P500 index during the economic downturn of 2008.
We will first go through some of the math used to develop the Breakout Detection algorithm, which makes use of some techniques from the areas of statistics known as Robust Statistics and Energy Statistics. Next we will see how to install and implement the package, and finally look at some examples of this package applied to some time series data.