Neurophysiological computational five-factor modeling of affect for analyzing human behavior

Abstract

There is a general consensus among academics in the fields of Psychology and Behavioral Sciences that people are inherently different and they handle similar jobs in different ways. That is because each individual has unique characteristic patterns of personality traits that modulate and control his/her behavior. However, someone’s behavior can be predicted by assessing his/her personality traits. Therefore, personality tests have become increasingly popular in corporate companies to aid in hiring new employees. In addition, personality assessments have been included in the field of computer science; particularly to support human-computer interactions. The traditional way of measuring the personality has been through self-reported instruments such as interviews and questionnaires. Such measurements are dependent on human responses and behavior which are influenced by environment, subjectivity and cultural biasness. Studies have reported that behavior is highly influenced by emotions and have suggested that behavior is caused by emotions. On the other hand, studies have reported strong qualitative correlations between personality dimensions and emotions. Emotions as mental states can be measured from spontaneous brain signals using an EEG-based emotion recognition system. Therefore, the main objective of this research is to quantify the correlation between emotions and personality to formulate a computational model that can be used for measuring personality traits based on its correlation with EEG-based affect, particularly with emotion primitives; valence and arousal. Such measurements will be based on spontaneous brain signals, and thereby will be independent of undesirable intended behavior such as misremembering, lying and cultural biasness. To that end, 22 subjects have participated in the experiments. Their personality factors were measured by using the NEO-FFI questionnaire and their emotions were measured using EEG-based emotion recognition system while being exposed to four different emotional stimuli: happy, fear, sad and calm. The relationships between each personality factor and emotion primitives were both tested and quantified using Pearson’s correlation. Whenever the correlation is significant, the relationship is analyzed using a linear regression for modeling the relationship and extracting the prediction equation of that factor. Finally, the model was tested using emotion primitives as predictors on each participant to predict his/her personality. The findings have shown the ability of the model in measuring personality traits from brain signals, and that the model has high potential to assess four factors of personality with the best average error rates of 10.18%, 4.55%, 13.64%, and 8.09% for Neuroticism, Extraversion, Openness and Conscientiousness respectively. Even though, the correlation between the fifth factor (agreeableness) and emotion primitives for the used sample was not significant enough to be used for predicting the factor agreeableness.