Research on emotion recognition of autistic children based on electrocutaneous signals and electrocardiogram signals

Aiming at the problem that explicit emotional characteristics and a single physiological modality are not enough to support robust emotion recognition in autistic children, this paper proposes a random forest emotion classification algorithm based on implicit measurements (electrodermal signals and electrocardiogram signals). Through artificial induction, the physiological signals of the four emotions (happy, sad, scared, and angry) of the subjects were collected. Gaussian smoothing was performed on the electrocutaneous signals, and wavelet noise reduction processing and feature extraction were performed on the electrocardiographic signals to obtain physiological emotional indicators. Five types of machines were used to obtain the physiological emotional indicators. Learning methods perform sentiment classification on sentiment indicators, in which random forest shows good classification accuracy. Then, based on the grid search algorithm of the sklearn library, different combinations of key parameters of the random forest classifier were performed, and the optimal parameter combination was determined through five-fold cross-validation, and the ROC curve was used to measure its performance. The results show that the dual-modal emotion recognition based on random forest can classify four emotions with a maximum accuracy of 92.86%.