A hybrid neuro-experimental decision support system to classify overconfidence and performance in a simulated bubble using a passive BCI

Abstract

Significant advancements in brain-computer interfaces (BCIs) can lead to the development of enhanced decision-making platforms. Irrational behavior generating potential negative consequences can be better calibrated using experimental approaches. One such example is asset bubbles. Experimental methods using neuroimaging equipment can improve understanding of the phenomenon, while advancements in artificial intelligence can be used to classify and predict behavior based on the cognitive neurodynamics at play. Towards this goal, we ran an EEG-based experiment where subjects traded in a risky environment simulating the boom and bust of a financial bubble. We used blind source separation and empirical mode decomposition on EEG neural data to estimate approximate entropy as a metric of brain activity, which was subsequently used as feature vector input into various machine learning classifiers. We thus develop a decision support system to classify subjects according to overconfidence and trading performance using a hybrid classification methodology. We report average accuracy ratios ranging between 80 and 85%, using classifiers such as linear discriminants, decision trees, support vector machines, Naïve Bayes, multi-layer perceptron, and an artificial neural network. Not only does the hybrid method we employ produce high classification accuracy ratios in a cognitively demanding task of financial decision-making, but it can also be used in the classification of motor imagery tasks that can be coupled with financial decisions in the future. We contribute to the literature in neurofinance from an empirical perspective, while also providing a technical framework that can act as a basis for developing expert BCI systems.