Machine Learning Techniques for JetMET Data Certification of the CMS Detector

In CMS, data quality monitoring (DQM) and data certification (DC) are crucial components in ensuring reliable data quality suitable for physics analysis. In the offline DQM procedure, the quality of recorded data, grouped in "runs", is evaluated. The current method for certification of quantities related to hadronic jets and missing transverse momentum (MET) is mostly reliant on manually monitoring reference histograms summarizing the status and performance of the detector. Given the large number of distributions that are mentioned, the process is time intensive and prone to human error when deviations from the norm are less evident. The results presented here show machine learning methods for certifying offline DQM data, focusing on hadronic jet and MET objects. Using collision data collected during 2018, we show that autoencoder techniques can accurately certify runs and detect ineffective detector regions, allowing to reduce both the time required for DC, as well as the rate of anomalies a human expert can potentially miss.