Multiparameter Analysis of the Belle II Pixeldetector’s Data Using Principal Components Analysis and Neural Networks

Sumitted to PubDB: 2019-08-07

Category: Bachelor Thesis, Visibility: Public

Tags: -

Authors Stephanie Kaes, Jens Lange
Date Jan. 1, 2019
Belle II Number BELLE2-UTHESIS-2019-001
Abstract In a recent thesis, the use of self-organising maps (SOM) and feed forward networks for classification processes was applied in the analysis of data from Belle II’s pixel vertex detector. This thesis continues the work by examining correlations between the charge-related and size-related pixel cluster properties. To find correlations, principal components analysis was applied to a data set of antideuterons and beam background. Additionally, first attempts in the analysis of cluster shapes were made, observing that all clusters are either orientated in 45 ◦ direction or parallel to a module’s lateral borders. More than half of them can be grouped into single pixel clusters, rectangular, or quadratic shapes. The PCA-transformed data set was used to train a SOM. A comparison of the classification results to a SOM trained with the original data set lead to the result, that PCA does not necessarily improve the outcome. Lastly, use of a SOM was made to differentiate more than one particle from background, for the first time.
Conference Giessen

Files