Outlier removal of ATLAS online trigger rate data using isolation forests and local outlier detection

The ATLAS trigger is a combined hardware/software system designed to filter and record significant collisions. The number of events saved per second, or trigger rate, is often used for rapid decisions as experimental conditions evolve. Typically, these decisions are based on detector conditions information, which is subject to noise that leads to significant outliers. The goal of this project was to develop a method that would remove these unnecessary outliers. The method implemented combines Isolation Forest and Local Outlier Factor from Sci-kit Learn. Using this method we were able to remove significant noise from online trigger rate data.