Emotion Recognition Related to Stock Trading Using Machine Learning Algorithms With Feature Selection

Abstract

This article proposes an emotion elicitation method to develop our Stock-Emotion dataset: a collection of the participants' electroencephalogram (EEG) signals who paper-traded using real stock market data, virtual money, and outcomes that emotionally affected them. A system for emotion recognition using this dataset was tested. The system extracted from the EEG signals the following features: five frequency bands, Differential Entropy (DE), Differential Asymmetry (DASM), and Rational Asymmetry (RASM), for each band. Our system then carried out feature selection using a filter method (Mutual Information Matrix), combined with a wrapper process (Chi-Square statistics) and alternatively using the embedded algorithms in a Deep Learning classifier. Finally, this work classified emotions in four quadrants of the circumplex model using Random Forest and Deep Learning algorithms. Our findings show that 1) the proposed emotion elicitation method is useful to provoke affective states associated with trading, 2) the proposed feature selection process improved the classification performance of our emotion recognition system, and 3) classifier performance of the system can recognize trading related emotions and has results comparable with the state of the art research corresponding to a similar number of output classes.