Estimação de densidades multivariadas para a filtragem de eventos baseada em um detector de altas energias com fina segmentação

Abstract

The electron identification is of fundamental importance to the ATLAS physics pro-gram, at CERN. This Master's Thesis planned to study and to reproduce one of the main offline algorithms, based on nonparametric maximum likelihood estimation, ap-plied to identify electron/positron particles using the ATLAS Detector to then propose additional processing techniques that could improve its performance. The ATLAS Col-laboration simplifies the likelihood method by considering independence between the discriminant variables. This approach opens possibilities for improving the method by means of applying techniques capable of mitigating the variables dependence. Our main contribution lies in the implementation of an algorithm based on Multivariate Kernel Density Estimation (MKDE). This algorithm should be able to decrease the error caused by the variables dependence, as mentioned above, improving the ATLAS electron identification performance. The impact of this new proposal was also compa-red to the most used algorithms developed by ATLAS group, known as Egamma and Likelihood.