A random forest algorithm-based emotion recognition model for eye features

Abstract

Objective: To develop a random forest algorithm-based model for the recognition of angry, neutral, and happy emotions for eye features and to further analyze the importance of eye features. Method: Raw data were obtained using emotional images from the Chinese Emotional Face System (CAFPS), and the code was used to derive relevant eye features data to build the database. The relevant features were left and right pupil size, left and right visible iris size, Distance between inner corners of eyes, upper and lower eyelid distance, left eye opening and closing, AU1 (inner eyebrow raised), AU2 (outer eyebrow raised), AU4 (overall lowered eyebrow), AU5 (raised upper eyelid), AU6 (raised cheek) and AU7 (eye constriction), a total of 13 eye features, were used to construct an emotion recognition model using the random forest algorithm and to analyze the importance of the features. Results: The differences were statistically significant (p<0.01) in all 13 eye features; the accuracy of the model constructed using the random forest algorithm was 70.2%, the recall was 0.702, the accuracy was 0.977 and the F1 was 0.809. AU6 had the highest importance in the process of constructing the model, accounting for 15.4%. Conclusion: Eye features have a role in the process of building an emotion recognition model, validating the theories related to Chinese medicine eye diagnosis, and combining Chinese medicine eye diagnosis with theories related to Chinese medicine emotions to identify patients' emotions by capturing eye information, which has clinical practice implications.