Analysis and Assessment of Dynamics of Neurocomputer Performance Measures

Abstract

The article discusses the benefits of neurocomputer interfaces aimed at classifying patterns of EEG signals which are converted into output signals for controlling actuators. To detect and classify signals, artificial neural networks of various architectures are used to increase recognition efficiency. The accuracy of the biological signals depends on the condition how measurements were performed and filtered to prepare the data for training and testing of ANNs. The development of mathematical models and teaching methodology will facilitate neural networks testing of various types and configurations. It will also help to evaluate and improve the accuracy of the classifier as well as to eliminate and compensate for the shortcomings of various software products and libraries. The assessment of neurocomputer performance measures, even of a well-trained network implemented in the classifier, is complicated by identifying events in a dynamic measurement system. The creation of a system of adaptive filters and convolution of input signals made it possible to identify the necessary patterns in the EEG to facilitate the implementation of the neurocomputer interface.