A Comprehensive Analysis of the Success of Classification Algorithms for the Classification of Emotional States Based on the User's Behavioral Characteristics

Abstract

Behavioral characteristics are currently one of the most widely used ways how we can determine a user's emotional state. Authors in the contribution measures the possibilities of classifying the emotional state of the user based on the behavior when working with a computer and with a smartphone. The contribution follow to the results of the original work of the authors, which dealt with the development of the Emotnizer application, data collection and their processing and classification of the four basic emotions found simultaneously in Ekman and Russell's emotional model. The application has been completely redesigned and currently allows the classification of four basic dimensions (anger, joy, sadness and relaxation), while within these dimensions it is able to recognize and classify 20 different emotional states based on valence and arousal. In the article, we compare the results achieved by us (achieved by the 3-fold cross-validation method) with the results of the authors' most similar works from the analysis of the current state. We achieve the highest classification accuracy of up to 87.88% by classifying arousal (low, medium, high) with a neural network algorithm for the second data set. Among the classification results for valence (negative, neutral, positive), we proposed a neural network classifier that works with an accuracy of 85.61%.