SERO

Although our analysis was overarchingly accurate (70.17%), it is important to consider associated errors with the data collected, methodology and devices used. Inaccurate readings and self reporting within the data lended itself to complexity.

For example, there were readings with moderate accelerometer activity alongside high GPS speed, but the study participant reported the trip mode as being "stationary." This is a clear inaccuracy within the data readings collected through the accelerometer and GPS. Our methodology and attempt to classify activity type into four categories (Activity in Constrained Area, Sedentary Transportation, Active Transportation, Stationary) would have inaccurately classified the activity in such an instance.

Self-reporting was another issue. The methodology we utilized to validate our activity predictions was what the participant self-reported for their trip mode. Modes of transportation in this column includes phrases such as "scooter," "pedestrian," and "vehicle." This may have lead to perpetuated inaccuracies within our study, as it was shown that in some cases this data related to how an individual previously arrived at a location, rather than what they were doing at that particular location during that time when compared with the column describing activity. We had chosen to use the column describing trip mode as opposed to the column describing activity because there were millions of entries and it would have been time-consuming to classify the responses of each study participant in the highly-variable column describing activity.

Although the methods of trajectory data masking we explored in our literature review were unavailable to us due to restraints of our lab (software and computing power), we were able to ensure k-anonymity through geovisualization techniques (uni representation of trajectories, lack of accurate base map, and low resolution rasterization).

While we were impressed with the results obtained, we understand there are some areas for improvement that can be worked upon in future studies. The findings of this paper demonstrates the relationship between accelerometry and GPS, especially with regard to activity prediction. Future studies should attempt to account for the discrepancies between the accelerometer and GPS data and the self-reported trip mode column or the activity description column.