An Efficient Approach to Mental Sentiment Classification with EEG-based Signals Using LSTM Neural Network

This research explores the prominent signals and presents an effective approach to identify emotional experiences and mental states based on EEG signals. First, PCA is used to reduce the data’s dimensionality from 2K and 1K down to 10 and 15 while improving the performance. Then, regarding the insufficient high-quality training data for building EEG-based recognition methods, a multi-generator conditional GAN is presented for the generation of high-quality artificial data that covers a more complete distribution of actual data by utilizing different generators. Finally, to perform classification, a new hybrid LSTM-SVM model is introduced. The proposed hybrid network attained overall accuracy of 99.43% in EEG emotion state classification and showed an outstanding performance in identifying the mental states with accuracy of 99.27%. The introduced approach successfully combines two prominent targets of machine learning: high accuracy and small feature size, and demonstrates a great potential to be utilized in future classification tasks.