An Efficient Analysis of EEG Signals to Perform Emotion Analysis

The analysis of human emotional features is a significant hurdle to surmount on the path to understanding the human mind. Human emotions are convoluted thus making its analysis even more daunting. In this paper, a meticulous and thorough analysis of EEG Brainwave Dataset: Feeling Emotions is performed in order to classify three basic sentiments experienced by people. A Machine Learning (ML) based framework is proposed to execute a multi-class classification process to identify positive, neutral and negative emotional experiences in people. The dataset is analysed in two distinct ways. The first method employs chi-square algorithm to select 500 of the best features from each sample in the dataset which are then employed in classifying multiple emotions utilizing several machine learning models. The second method utilizes the sparsePCA algorithm for feature extraction before conducting a multi-class classification with the help of machine learning models. It is supplemented with a binary classification process of each of the individual sentiments available in the entire dataset to analyze the efficacy of these ML models. The ML algorithms-Support Vector Machines (SVM), Random Forest (RF), Light Gradient Boosting Machines (LGBM) and Multi-Layer Perceptron (MLP) are employed in this investigative study. Maximum accuracy of 99.25% and a precision of 99.25% is obtained from the application of the LGBM model after the optimization of hyper-parameters after sparsePCA is used in feature extraction.