Explainable AI for real time micro expression emotion detection

Abstract

Human emotions are frequently used in a variety of domains to decide how best to handle distinct clientele. One of the ways these emotions can be expressed is using facial expressions and this is particularly useful for those who are unable to express their emotions verbally for a variety of reasons. In an attempt to do accurate facial emotion recognition, new interest has been shown in micro-expression detection—the capacity to recognize transient facial expressions that convey genuine emotions. This has often been carried out using Artificial Intelligence and particularly deep learning techniques which are used to train models to handle this task . However, the deep learning models’ interpretability and general adoption are sometimes hampered by their complexity and black-box nature. To address this challenge, we applied different explainability techniques on various models but particularly Local Interpretable Model-agnostic Explanations (LIME). By enforcing LIME and the other explainability techniques, we were able to get insights into the decision making process and hence enhanced the transparency of our models. Using a dataset of eight emotion classes, we show the effectiveness of our method and highlight LIME’s capacity to explain the elements and patterns that influence model choices. This improved interpretability opens the door to more dependable and trustworthy micro-expression detection systems with less bias and more equity, which will promote wider adoption and better applications in the fields of security, healthcare, and human-computer interaction.