For the second-level triggering system of the ATLAS detector in the LHC experiment, calorimeters play a major role. The calorimetry system for ATLAS comprises highly segmented electromagnetic and hadronic sections, which provide a detailed energy deposition profile for interactions. For designing efficient and fast triggering schemes based on calorimetry, preprocessing methods have to be developed for reducing significantly the high dimensionality of the original input data space (496 components, for the simulation data used in this work), while retaining the most important features measured by the detector. For such data compaction, principal component analysis is well known to be an efficient method. Here, principal components are extracted from each individual calorimeter segment of both electromagnetic and hadronic calorimeter parts. This component extraction procedure aims to better reveal the subtle differences that arise in energy deposition profiles for electrons and jets that passed the conditions of the efficient first-level triggering system of ATLAS. Only twenty-two components for the sections of the electromagnetic part and eleven for the hadronic ones suffice to account for most of data variance. Original data projected on these 33 components are fed into a neural classifier trained with backpropagation method, which achieves 98.4% electron efficiency with less than 2.9% of jets being misclassified as electrons. This result is compared with that obtained with different topological methods previously developed for the triggering task and visualization of the information being extracted by such principal components points out from where comes the most relevant information identified by the neural discriminator.