One of the most effective tools for predicting time series is the use of machine learning methods, in particular, artificial neural networks. However, at the same time, a necessary stage of the study is to reduce the dimension of the input data. In this paper, we consider the results of data dimensionality reduction based on the ranking of input features by their significance in solving the problem of forecasting the geomagnetic Dst index. To assess the relative significance of features, an iterative approach is used, associated with the search of candidate models by discarding features one by one based on simple linear regression models.

There is a question about the appropriate emotional and expressive language of a virtual actor. In this paper we study facial expressions. We investigate the transformations between the space of Action Units and the standard affective space in the loop of nonverbal communication between a person and a virtual actor using facial expressions [1]. We are mapping both dimensions into each other using various machine learning algorithms. Action Units space was mapped into emotional space directly using artificial neural networks. Emotional space was mapped into Action Units space with help of dimensionality reduction followed by clusterization of the latter. After the final synthesis, the facial expression of virtual actor can be determined.

This paper performs comparative analysis of the prediction quality for the time series of hourly average flux of relativistic electrons (E > 2 meV) in the near-Earth space 1 to 24 h ahead by various machine learning methods. As inputs, all the predictive models used hourly average values of the parameters of solar wind and interplanetary magnetic field measured at the Lagrange point between the Sun and the Earth, the values of Dst and Kp geomagnetic indexes, and the values of the flux of relativistic electrons itself. The machine learning methods used for prediction were the multi-layer perceptron (MLP) type artificial neural networks, the decision tree (random forest) method, and gradient boosting. A comparison of the quality indicators of short-term forecasts with a horizon of one to 24 h showed that the best results were demonstrated by the MLP. The horizon of satisfactory forecast accuracy on independent data is 9 h, the horizon of acceptable accuracy is 12 h.