EEG Multiclass Emotion Recognition

Abstract

"Electroencephalography (EEG) is a promising way to study brain activity and is being used to figure out how people’s emotions work, which is important in fields like neurological and mental health examinations. But it's hard for standard machine learning methods to understand the complicated, high-dimensional data that EEG signals. This research looks at how well a mechanism that combines Long Short Term Memory (LSTMs) Convolutional Neural Networks (CNNs) can help classify emotions from EEG data. The author was aware of the problems with only using CNN-based methods, especially when dealing with the changing timing of EEG data. To solve these problems, the author came up with a novel approach that uses CNNs to pull out spatial features and LSTMs to record the temporal sequences that go along with changes in emotions. When these architectures work together, they try to describe both the location and timing of brain activity that is connected to emotional states. Used benchmark EEG emotion dataset to test our model and see how well it did compared to normal CNN and LSTM models. Our tests showed that the hybrid model did better than these traditional methods, showing more accurate mood classification. Also saw that fine-tuning the hyperparameters could make the model much better at making predictions. Our findings show that combining CNNs and LSTMs has a lot of potential for making EEG-based mood classification systems that are more accurate and reliable. This study adds to the current conversation in affective computing and makes the groundwork for further progress in the area in the future. Keywords: EEG, Emotion Classification, Convolutional Neural Networks, Long Short Term Memory, Affective Computing Subject Descriptors: • Neuroscience → Affective Computing → EEG Analysis Computer Science → Machine Learning → Deep Learning → Hybrid Model"