Convolutional neural network-based emotion recognition using recursive feature elimination

Emotion detection plays a crucial role in fields such as biomedical applications, smart environments, brain-computer interfaces, communication, security, and safe driving. In this paper, we present a novel approach for detecting emotions using electroencephalogram signals. The method employs convolutional neural network (CNN) as the classifier, which is chosen from a variety of intelligent algorithms. Discrete wavelet transform is used to decompose the signals into four frequency bands including theta, alpha, beta, and gamma. These bands are then utilized for feature extraction. Out of a total of 1920 features, the recursive feature elimination algorithm based on random forest model combining with 5-fold cross-validation and the K-nearest neighbors model, selects the 720 most relevant features. The proposed algorithm is further validated on the selected feature subset using 5-fold cross-validation with CNN on the validation set. The results demonstrate the potential of this algorithm for emotion recognition.