EVALUATION OF HUMAN PSYCHOLOGICAL STATE USING MLP NEURAL NETWORKS

Abstract

This article provides an overview of the construction of an MLP neural network for estimation of a person's psychological state. The problem of critical sessment and direct psychological help is quite important nowadays, as most people are faced with problems, both personal and national or global, such as political instability or a pandemic. Most of the existing systems for assessing the human condition are aimed at helping solve problems related to physical health. A smaller part of similar systems is intended to work with a mental state, however, they are more highly specialized and directed to work with an already known problem. A dataset was formed based on people's responses regarding their well-being. The MLP type of network was chosen, since this type is quite suitable for the given classification task. Three types of models are considered: basic MLP, MLP with ReLU activation and Unet-like model. The process of selecting the optimization algorithm and loss function is described. The article shows an overview and assessment of training effectiveness for each of the chosen models. Accuracy on a test set is shown. In addition, a description of actions related to attempts to improve the accuracy of the network (changing the number of questions, normalizing the initial data) is provided. A description of possible algorithms for data normalization is provided. In general, this article reveals a possible approach to the construction of a neural network, which can be useful not only for assessing one's own psychological state, but also for specialists working in the field of psychology, since they will be able to use a similar network to assess a person's state or compare their own assessment with an assessment system thereby increasing the accuracy of assessment.