Abstract
Humans use facial expressions in many contexts to communicate their ideas or weigh their emotions. Deaf people depend on these expressions mainly in daily communications. They use the facial expressions to add the grammatical meaning for sentences of similar words. Therefore, developing smart systems to recognize facial expressions becomes a necessity. The main obstacle comes from the uncertainty and ambiguity of grammatical facial decisions. Hence, fuzzy and fuzzy rough artificial intelligent algorithms formulate feasible solutions to make decisions in such situations. This paper presents a hybrid of fuzzy rough feature selection inspired by ANT Colony Optimization (FRFS-ACO) and fuzzy rough nearest neighbor (FRNN) classification algorithms to decide about different facial expressions. The proposed hybrid is compared to other artificial algorithms and hybrids to judge its accuracy and efficiency. The experiments are accomplished using a standard grammatical facial expressions data set with nine different emotions recorded by Microsoft Kinect sensor and kept on the UCI machine learning repository. The experiments and comparisons clarified that the proposed hybrid provide feasible average accuracy level of 93.7% and dramatically reduced the required classification time.