Learning bayesian network to predict group emotion in kindergarten by evolutionary computation

Abstract

In the educational services, students’ emotions are an important factor that determine its effect. We have previously conducted research that led them to target emotions using environmental factors. However, the study used the bayesian network based on domain knowledge to predict emotions, which may differ from the actual environment. In this paper, we propose a method to learn the bayesian network for group emotion prediction in kindergarten from data through evolutionary computation. The learning data are brightness, color temperature, sound, volume, smell, temperature, humidity, and current emotion. The structure of the network is encoded with two chromosomes to represent nodes and arcs. To explore the optimal structure, evolutionary operators are used that can convey information in sets. We also experiment with various inference nodes not observed. Experimental results show that the accuracy is 85% with 20 inference nodes, which can replace network designed with domain knowledge. By comparing the evolution of the best model, we analyze the influential factors that determine the structure.