Deep Neural Networks Applied to User Recognition Based on Keystroke Dynamics: Learning from Raw Data.

Several studies have investigated how to use Machine Learning algorithms to recognize users based on keystroke dynamic. All those studies required Feature Engineering (FE), i.e., a process in which specialists choose what attributes should be considered for learning. However, this process is susceptible to problems such as original information loss or inappropriate attribute choices. Thus, the objective of this work is to demonstrate the hypothesis that user recognition algorithms applied to keystroke dynamics raw (original) data can perform better than the ones that depend on FE. Therefore, this work proposes a deep neural network named DRK. The proposed network contains layers that learn adequate data representations to perform user recognition based on keystroke dynamics raw data, avoiding FE. Experiments compared DRK with four other deep neural networks that use FE in four datasets with 280 users. The proposed network achieved better results in all datasets, showing strong evidence that the stated hypothesis is, in fact, valid.