Optimized Emotion Recognition Using Swarm Intelligence and Explainable AI for EEG Based Analysis

The identification of the emotional arousal and valence states of the people has been a major concern in various disciplines, includingthe adaptation of social life to differently able people. Automated identification systems for such purposes can not rely solely onbasic postures and facial gestures alone. Advanced methodologies, such as neural signal analysis, can complement conventionalmethodologies by using EEG data with a view to understanding emotional states and cognitive processes. In order to develop arobust system, in the current work, we utilized the GAMEEMO dataset, which has the advantage of combining the flexibility ofincorporating various game environment variables and coupling them with corresponding psychological state labels. For the purposeof this study, two emotions are considered for the classification: Positive and Negative emotions. Acquiring EEG signals from thebrain underwent some pre-processing techniques such as bandpass filter and wavelet transform is sensitive to temporal frequencyvariations. Subsequently, feature extraction is performed using the Fast Fourier Transform from which vectors were extracted.Finally, the extracted time-frequency features were applied to a machine learning technique called Support Vector Machines, havinga particle swam optimization classifier for emotion classification. Our study considers multiple emotions, specifically ’Arousal’ and’Valence’ for the precision of classification of LANV, LAPV, HANV and HAPV. These experimental results represent the completework, proving the significance and efficiency of the proposed SVM with the PSO achieved 99.37% of the accuracy of emotiondetection. In addition, our study used explainable artificial intelligence methods like SHAP and LIME to determine which featuresare the most important to improve the accuracy of EEG-based emotion classifiers.