Reward-penalty weighted ensemble for emotion state classification from multi-modal data streams

Abstract

Researchers have shown the limitations of using the single-modal data stream for emotion classification. Multi-modal data streams are therefore deemed necessary to improve the accuracy and performance of online emotion classifiers. An online decision ensemble is a widely used approach to classify emotions in real-time using multi-modal data streams. There is a plethora of online ensemble approaches; these approaches use a fixed parameter(ß) to adjust the weights of each classifier (called penalty) in case of wrong classification and no reward for a good performing classifier. Also, the performance of the ensemble depends on the ß, which is set using trial and error. This paper presents a new Reward Penaltybased Weighted Ensemble (RPWE) for real-time multi-modal emotion classification using multi-modal physiological data streams. The proposed RPWE is thoroughly tested using two prevalent benchmark data sets, DEAP and AMIGOS. The first experiment confirms the impact of the base stream classifier with RPWE for emotion classification in real-time. The RPWE is compared with different popular and widely used online ensemble approaches using multi-modal data streams in the second experiment. The average balanced accuracy, F1-score results showed the usefulness and robustness of RPWE in emotion classification in real-time from the multi-modal data stream.

Arijit Nandi is a fellow of Eurecat’s “Vicente López” PhD grant program. This study has been partially funded by ACCIO, Spain (Pla d’Actuació de Centres Tecnològics 2021) under the project TutorIA.

Peer Reviewed

Postprint (author's final draft)