Innovative Measures of Verhulst Diagram for Emotion Recognition using Eye-Blinking Variability

Background: The human body continuously reveals the status of several organs through biomedical signals. Over time, biomedical signal acquisition, monitoring, and analysis have captured the attention of many scientists for further prediction, diagnosis, decision-making, and recognition. Recently, building an intelligent emotion recognition system has become a challenging issue using the application of signal processing. Frequently, human emotion classification was proposed utilizing the internal body status in dealing with affective provocations. However, external states, such as eye movements, have been claimed to convey practical information about the participant’s emotions. In this study, we proposed an automatic emotion recognition scheme through the analysis of a single-modal eye-blinking variability. Methods: Initially, the signal was transformed into a 2D space using the Verhulst diagram, a simple analysis based on the signal’s dynamics. Next, some innovative features were introduced to characterize the maps. Then, the extracted measures were inputted to the support vector machine (SVM) and k-nearest neighbor (kNN). The former classifier was evaluated with three kernel functions, including RBF, linear, and polynomial. The latter performances were examined with different values for k. Moreover, the classification results were assessed in two feature-set partitioning modes: a 5-fold and 10-fold cross-validation. Results: The results showed a statistically significant difference between neutral/fear and neutral/sadness for all Verhulst indices. Also, the average values of these characteristics were higher for fear and sadness than those of other emotions. Our results indicated a maximum rate of 100% for the fear/neutral classification. Therefore, the suggested Verhulst-based approach was supremely talented in emotion classification and analysis using eye-blinking signals. Conclusion: The novel biomarkers set the scene for designing a simple accurate emotion recognition system. Additionally, this experiment could fortify the territory of ocular affective computing, and open a new horizon for diagnosing or treating various emotion deficiency disorders.