A segmented principal component analysis applied to calorimetry information at ATLAS

A segmented principal component analysis is applied for dimensionality reduction of the calorimeter information at the second level trigger of ATLAS. The segmented analysis is proposed in order to fully explore the high segmentation of the calorimeter system and the different levels of granularity present at each segment of the hadronic and electromagnetic sections. Considering electron and jet simulated events, a high data compaction level (above 96%) is achieved, even when preserving 95% of the original data variance. Using data projection onto the principal components of each calorimeter segment, and a neural classifier, 97.3% of electrons are correctly identified for a misclassification of jets below 9%.