Cloud-Based Human Emotion Classification Model from EEG Signals

Human emotions are complex, mutiifaceted phenomena, defined as multidimensional subjective experiences influenced by several factors like cognitive processing, social norms, genetics etc. Emotion recognition can be stated as a vital aspect of artificial intelligence which is intensively utilized in the enhancement of human-brain-computer interface in recent days. The complexity of emotional analysis can be dealt with in depth by classifying and analyzing the physiological signals over non-physiological signals. A potential non-invasive technique for measuring the electrical activities of the human brain technology is Electroencephalography (EEG). In response to different stimuli, the event-related-potentials can be measured by EEG to classify different emotional states with the help of machine learning algorithms. This study focuses on building machine learning models (Logistic Regression, Decision Forest, and Boosted Decision Tree) for classifying emotional stages (positive, negative, and neutral) on a cloud-based environment, Microsoft Azure Cloud. It concludes by comparing the classification models by gauging the evaluation metrics, i.e., precision, recall, and accuracy. The highest accuracy is achieved by the Boosted Decision Tree model which is around 99% and this has surpassed the performance of other existing state-of-art models for human emotion classification. The obtained accuracy for Decision Forest and Logistic Regression are consecutively 96% and 93%. To generate an optimal result and avoid overfitting/over-learning issues, some remarkable measures have been undertaken, i.e., tuning and iterating of hyperparameters, optimizing tree depth and nodes numbers, controlling random seeds etc. Our valuable findings will provide significant insights into psychological research to collaborate human-brain-computer interface with artificial intelligence.