Predicting group emotion in kindergarten classes by modular Bayesian networks

Conventional methods predict emotion directly by measuring equipment like electrode. However, this approach is not suitable for education, especially for children. In this paper, we propose modular Bayesian networks for predicting the emotion with the environment information from the sensors. The Bayesian network is constructed as modules divided by Markov boundary. To evaluate the proposed method, we use data collected from kindergarten classes. The results show more than 84% accuracy and 20 times faster than the single Bayesian network.