Evaluating the Accuracy of Wearable Devices for Positioning Classification

Satellite-based positioning is among the useful functionalities provided by wearable devices. In active and assisted living, positioning information from wearable devices may be exploited to support subjects affected by cognitive impairments, but it suffers from a limited measurement accuracy. This paper presents a method based on Machine Learning (ML) for positioning classification. Differently from other ML algorithms, the method here proposed leverages the measurement error committed by wearable devices in acquiring coordinates of geographical positions. Then, the tests are performed on three commercial smartwatches, in two near outdoor positions with different conditions of non-optimal signal reception. The proposed method generally exhibits notable performance, reaching on average a classification accuracy of 91.97%.