Classifier Fusion Method Based Emotion Recognition for Mobile Phone Users

With the development of modern society, people are paying more and more attention to their mental situation. An emotion is an external reaction of people’s psychological state. Therefore, emotion recognition has attached widespread attention and become a hot research topic. Currently, researchers identify people’s emotion mainly based on their facial expression, human behavior, physiological signals, etc. These traditional methods usually require some additional ancillary equipment to obtain information. This always inevitably makes trouble for users. At the same time, ordinary smart-phones are equipped with a lot of sensor devices nowadays. This enables researchers to collect emotion-related information of mobile users just using their mobile phones. In this paper, we track daily behavior data of 50 student volunteers using sensors on their smart-phones. Then a machine learning based classifier pool is constructed with considering diversity and complementary. Base classifiers with high inconsistent are combined using a dynamic adaptive fusion strategy. The weights of base classifiers are learned based on their prior probabilities and class-conditional probabilities. Finally, the emotion status of mobile phone users are predicted.