In this paper, we present an approach to improve the accuracy of hand tremor severity in Parkinson's patients in real-life unconstrained environments. The system leverages data achieved from daily interaction people with their smartphones and uses technologies for classifying and combining data. We describe the basic concept of data fusion and demonstrate how different combination techniques can improve the accuracy of tremor detection. The fusion enable to achieve the 23.5% improvement with respect to the average of individual classification models.
Keywords: Data fusion; Parkinson’s disease; hand tremor.