Advanced EEG Analysis Based Emotion Recognition: A Deep Learning Classifier with Hybrid Feature Selection and Artifact Reduction

Emotion recognition from Electroencephalogram (EEG) signals has emerged as a crucial area of research with a broad spectrum of applications. To enhance the accuracy and effectiveness of emotion classification, this study presents an innovative approach that combines advanced signal processing and deep learning techniques. The proposed methodology is structured into several distinct phases for optimal performance. Initially, EEG signals are preprocessed using a Fractional Order Butterworth (FOB) filter, which exhibits flexibility, allowing for precise control over the trade-off between preserving relevant emotional information and mitigating unwanted interference. The Short Time Fourier Transform (STFT) is applied to extract time-frequency representations from the preprocessed EEG signals. This transformation captures dynamic changes in spectral content, providing a comprehensive view of the emotional state over time. A hybrid approach is employed to optimize the feature set and enhance the efficiency of emotion classification. This approach combines the Improved Artificial Fish Swarm algorithm with Particle Swarm Optimization (IAFS-PSO). Combining these algorithms efficiently navigates the solution space to select the most informative features, ensuring the subsequent classification process is based on a highly relevant and discriminative set of features. Emotion recognition is accomplished using a hybrid attention-based Convolutional Neural Network combined with Long Short-Term Memory (CNN-LSTM) networks. The CNN component captures spatial features in the EEG data, while the LSTM component is adapted to modelling temporal dependencies. The hybrid architecture is further enhanced with attention mechanisms, allowing the model to focus on critical segments of the EEG data, thereby improving classification accuracy. The evaluation of this approach is conducted rigorously, and the results of this study highlight the proposed methodology’s effectiveness and efficiency in accurately classifying emotional states from EEG signals.