Real-time emotion prediction system using big data analytics

Abstract

Emotions are an inseparable part of human existence. Emotions have a big impact on the success and failure of the human race. Comprehending human emotions could prove beneficial in creating improved systems for education, security, market sales, production, healthcare and other areas. Big data analytics applied to streamlined real time emotion sensor’s data can give new insights to anticipate stress before it arises and help in making significant choices that improve people's quality of life. This work proposes a framework for big data management and analysis of GSR sensor’s data in real-time for predicting emotions in human participants. Supervised learning techniques, Ensemble Boosted Tree, Neural Network, Naïve Bayes, Support Vector Machine, Decision Tree, K-nearest Neighbor, and Quadratic Discriminant Analysis are applied to the collected data. Two distinct criteria have been utilized for testing on real-time data one is trained on all participant data, resulting in a generalized system, while the other is trained on participant-specific data, resulting in a personalized system. Hence, the personalized system achieves an accuracy of up to 80.64% across all classes and 100% for binary classes as compare to generalized system achieves 78.12% accuracy. It is concluded that for the purpose of predicting emotions, the personalized model performs better than the generalized model.