Emotion analysis of EEG signals using proximity-conserving auto-encoder (PCAE) and ensemble techniques.

Emotion recognition plays a crucial role in brain-computer interfaces (BCI) which helps to identify and classify human emotions as positive, negative, and neutral. Emotion analysis in BCI maintains a substantial perspective in distinct fields such as healthcare, education, gaming, and human-computer interaction. In healthcare, emotion analysis based on electroencephalography (EEG) signals is deployed to provide personalized support for patients with autism or mood disorders. Recently, several deep learning (DL) based approaches have been developed for accurate emotion recognition tasks. Yet, previous works often struggle with poor recognition accuracy, high dimensionality, and high computational time. This research work designed an innovative framework named Proximity-conserving Auto-encoder (PCAE) for accurate emotion recognition based on EEG signals and resolves challenges faced by traditional emotion analysis techniques. For preserving local structures among the EEG data and reducing dimensionality, the proposed PCAE approach is introduced and it captures the essential features related to emotional states. The EEG data are collected from the EEG Brainwave dataset using a Muse EEG headband and applying preprocessing steps to enhance signal quality. The proposed PCAE model incorporates multiple convolution and deconvolution layers for encoding and decoding and deploys a Local Proximity Preservation Layer for preserving local correlations in the latent space. In addition, it develops a Proximity-conserving Squeeze-and-Excitation Auto-encoder (PC-SEAE) model to further improve the feature extraction ability of the PCAE technique. The proposed PCAE technique utilizes Maximum Mean Discrepancy (MMD) regularization to decrease the distribution discrepancy between input data and the extracted features. Moreover, the proposed model designs an ensemble model for emotion categorization that incorporates a one-versus-support vector machine (SVM), random forest (RF), and Long Short-Term Memory (LSTM) networks by utilizing each classifier's strength to enhance classification accuracy. Further, the performance of the proposed PCAE model is evaluated using diverse performance measures and the model attains outstanding results including accuracy, precision, and Kappa coefficient of 98.87%, 98.69%, and 0.983 respectively. This experimental validation proves that the proposed PCAE framework provides a significant contribution to accurate emotion recognition and classification systems.