Enhanced Mental State Classification Using EEG-Based Brain-Computer Interface Through Deep Learning

This study is divided into two main components. Firstly, it involves the design and training of multiple Convolutional Neural Networks (CNN) for the classification of brainwaves, predicting the mental state of an individual. Secondly, it encompasses the development of a Brain-Computer Interface (BCI) designed to record brainwaves, offering a user-friendly means to predict mental states using the recorded data and the trained neural network. The study utilizes a publicly available electroencephalographic (EEG) dataset collected with the Muse EEG headband. Various preprocessing techniques such as wavelet transform (WT), feature extraction, and feature selection are explored. The chosen temporal and statistical features are transformed into 2D grayscale images to facilitate the training of CNN models, classifying mental states into three categories: concentrated, neutral, and relaxed. The achieved highest accuracy is 91.72%, demonstrating competitiveness and improvement compared to previous works using the same dataset. The selected CNN model performs a fusion of the selected features and is integrated into the BCI, enabling users to predict mental states using EEG data. This BCI also holds the potential for enhancing model accuracy through continuous testing and incorporation of valuable data into the training dataset.