Multimodal Pain Level Recognition using Majority Voting Technique

Abstract

The measurement of subjective pain is still a problem especially with people who have verbal or cognitive impairments. In this work, we analyze the problem of some patients who did not express their pain through their facial muscles, but they were expressing it involuntarily through the autonomic neural system that can be observed in the physiological signals. An ensemble learning algorithm consisting of multimodal models trained on the geometric facial expressions and the physiological signals is proposed. Each model provides a certainty measure and the pain level is assigned by the most certain model. The proposed system is compared with the previous models.