Multi level SVM for subject independent agitation detection

The need to automate the detection of agitation for dementia patients is a major requirement for caregivers. This research aims at detecting the agitation status of the subjects using soft computing techniques that does not require supervision beyond the training phase. An autonomous multi-sensory device has been developed to achieve automatic assessment of agitation and to control stimulation that will reduce the agitation level automatically. The focus of this paper is the agitation detection algorithm. Three vital signs are monitored for agitation detection: the Heart Rate (HR) the Galvanic Skin Response (GSR) and Skin Temperature (ST). These measures are fed into a new SVM architecture: ldquoThe Multi level SVM learning machinerdquo. Results show very high detection accuracy of agitation, quick adaptation to the subject and a strong correlation between the physiological signals monitored and the emotional states of the subjects. The result is a learning algorithm that is ldquoSubject-Independentrdquo.