If one records speeds and distance, it happens that in (almost) closed spaces, the GPS signal jumps without impacting (for some reason) the XDOP precision. Those points could be filtered because the speed calculated by distance divided by time elapsed between 2 points is much higher than the speed reported by the GPS.
In the attached sheet, I\'ve calculated this speed as well as a factor between calculated and actual speed. You can see from the pivot table that this factor is very steep so that it could make a perfect criteria to filter out such wrong points.
Note: the ODS file is just an extract to show the formel (for file size reasons), but the CSV file represents the complete pivot showing the distribution of the factor by rounded values.