The Minimum Sampling Rate and Sampling Duration When Applying Geolocation Data Technology to Human Activity Monitoring

摘要

The availability of geolocation sensors embedded in smartphones introduces opportunities to monitor behaviours of individuals. However, sensing geolocation at high sampling rates can affect the battery life of smartphones. In this study, we sought to explore the minimum sampling rate of geolocation data required to accurately recognise out-of-home activities. We collected geolocation data from 19 volunteers sampled every 10 s for 8 non-consecutive days on average. These volunteers were also instructed to complete a paper-based activity diary to record all activities during each data collection day. For finding the minimum sampling rate, we derived datasets at lower sampling rates by down sampling the original data. A semantic analysis was applied using a previously published activity recognition algorithm. The impact of the sampling rates on accuracy of the algorithm was measured through the F_1 score. The best F_1 score was found at sampling intervals of 2 min and it did not drop substantially until the sampling intervals increased to 10 min. Our study proves the feasibility of monitoring activities at low sampling rates using smartphone-based geolocation sensing.