Given the extreme variations in people’s running, workout diversity, and the quality of consumer recording devices, run type classification is a complex (mathematical) problem – even we at TrainAsONE find it very difficult to classify some runs (and can get it wrong).
This problem falls under the realm of pattern recognition, and can be compared to handwriting or image recognition. Presently the system is around 80% accurate, though there are wide variations between individuals with some seeing much lower ‘hit rates’ and others seeing figures over 90%. As such, do not be surprised when the system ‘gets it wrong’. However, please ensure to correct the system using the run confirmation widget. In this way the system will continue to learn – a primary requisite of such AI is lots of data, and the better quality of the data, the better. In addition, with the algorithm enhancements we have planned, we are hoping to achieve over 95% across all users in the not too distant future.
It should also be said that the result of classification is only used during plan generation as part of a safety rule, and so not a major contributor to your plan. In time, it is our intention that run classification will become a purely informational tool, and bear no input on your plan.
Please see our ‘My run is classified incorrectly. How do I correct it? FAQ for further guidance.