Increasing the Validity and Trustworthiness of Automated Facial Emotion Recognition Systems

Since the inception of Affective Computing, there has been a meteoric rise in research surrounding the automation of human emotion detection and recognition using technology. Coupling with the growing trends of machine learning in recent years, there has been more research attention to build a valid and trustworthy emotion recognition system than ever across domains such as marketing, healthcare, education, safety and entertainment. Facial Emotion Recognition (FER) systems have been widely applied across the above-mentioned domains. However, current FER systems often lack validity and trustworthiness due to their reliance on static image analysis, disregarding the temporal evolution of emotions. This paper presents a novel approach that enhances the validity and trustworthiness of FER systems by incorporating temporal context through automated analysis of emotion time-series. By utilizing Long Short-Term Memory (LSTM) networks to process historical emotion predictions, the proposed method refines emotion classification through incorporating dynamic affective states over time as context. The findings of this paper contribute to the advancement of affective computing by introducing a more content-aware and resilient facial emotion recognition system.