Neurodynamic health diagnosis before, during and following space flights

The urgency is based on need developing algorithms for detecting health state of an astronaut. The main aim of the study lies in developing neural network model for breathing that will allow recognition of breath patterns and predicting anomalies that may occur. Class of machine learning algorithms includes many models, widespread feed forward networks are able to solve task of classification, but are not quite suitable for processing time-series data. The paper describes results of teaching and testing several types of dynamic or recurrent networks: NARX, Elman, distributed and focused time delay. The methods used in the study include machine learning algorithms, dynamic neural network architectures, focused time-delay network, distributed time-delay network, non-linear autoregressive exogenous model. Networks were built using Matlab Neural Network Toolbox 2014a software. For the purpose of research we used dataset that contained 39 polysomnographic recording. Records were obtained by pulmonology department of Third Tomsk City Hospital; on average recording were 8-10 hours long and included electrocardiography and oronasal airflow. Frequency of these signals was 11 Hz. The results include comparison of training and testing performances for various types of dynamic neural networks. According to classification accuracy obtained for learning and testing set of data, the most accurate results were achieved by non-linear autoregressive exogenous model.