Significant Parameters of the Keystroke for the Formation of the Input Field of a Convolutional Neural Network

The article deals with improvement of personality and emotion recognizers based on the keystroke pattern. The article shows that improvement of those recognizers can be implemented using neural network solutions. Based on the analysis of literature sources, we have determined the potential of using convolutional neural networks. The difficulties of using such networks are largely related to justification of a keystroke parameter list, whereby the input field of the convolutional neural network is determined. We suggest determining the said list with respect to feasibility of using the keystroke parameters that have a positive impact on efficiency of the neural network model. In addition, we assume that for admissible resource intensity the efficiency can be assessed experimentally using indices such as recognition accuracy or training and validation sample losses. The experiments allowed plotting the values of efficiency indices versus the number of training iterations for different options of keystroke parameters. The experimental results have shown that, for an input field formed using one keystroke parameter, the highest recognition accuracy is demonstrated by a convolutional neural network with the input field formed using the key hold time. At the same time, the combination of this parameter with other parameters does not result in any significant positive changes. We have justified the need for further research on improvement of keystroke parameters preprocessing procedures in order to increase their informational value. We have also determined that it is necessary to develop a method for determining the architectural parameters of a convolutional neural network designed to recognize the user personality and emotions based on their keystroke.