Research Effort in Emotion Recognition
Emotion is a complex subjective conscious experience that combines mental states, psychosomatic expressions and biological reactions of the body. The rapid techno- logical development of recent years combined with the ever-increasing interaction of man with computers may be the foundation for the development of systems and devices that can recognize, interpret and process human emotions.
The present master’s thesis examines the possibility of recognizing emotions through the Keystroke Dynamics, which refer to the analysis of data derived from the charac- teristics of a person’s typing. The TypeOfMood application for "smart" mobile phones with an iOS operating system is used as a basis for data collection, where the user declares his emotional and physical state through self-references.
The performance of the features is evaluated by three different classifiers: Logistic Regression, Support Vector Machines and Random Forest, in terms of F1-Score metrics and the Area under the Receiver Operating Characteristics curve (ROCAUC). It turns out that the Random Forest Classifier, using the Synthetic Minority Over- sampling Technique, achieves the best results for individual modeling per user for the "negative" emotional states ("Anxious" and "Sad"), while aggregate modeling on the whole has the best results for the "Happy" emotional state. However, a larger number of subjects and data is needed to generalize and verify the results.