Facial Expression Recognition (FER) programs often have difficulty correctly identifying the emotions of people of colour (POC). When compared to identifying the emotions of Caucasians, it is more often to see false positives or false negatives for POC. This is because there is a bias towards Caucasians in the AI system.
This poses a problem where accuracy is limited simply due to the lack of data available for POC. Thus, it is important to research ways to maximize the level of accuracy for identifying the emotions of POC to avoid false positives or false negatives when running a program.