I have two measuring devices. Both measure the same thing. One is accurate, the other is not, but does correlate with a non-fixed offset, some outliers, and some noise.

I won't always be using the accurate device. The nonfixed offset makes things difficult, but I'm certain there is sufficient similarity to make a link using a machine learning (or AI) technique and to convert one set of numbers to a good approximation of the other.

One is a footbed power meter and gives power in Watts every second. The other is a crank-based power meter, also outputting Watts at 1Hz. The footbed power is much less than the crank (which I know to be accurate), but it does track the increases and decreases in power, just with more noise and, as I say, a non-fixed offset (and by non-fixed I mean, at low power the offset is different to that at high power, I don't mean it isn't consistent, it is consistent). Both measure cadence which may be a useful metric to help find a pattern.

I will be collecting sets of data from both and hoped to plug the footbed data in as a column of values with the crank data as another column representing the truth, so after training, the model would be able to transform the footbed data to an approximation of the crank data.

Anyway, I'm completely lost as to how to begin. I've tried searching, but, clearly, I'm using the wrong keywords. Does anyone have any pointers, please?


1 Answer 1


OK, so I found the answer - it is to use multiple linear regression. I think this can be marked a solved, but I don't have enough rep to do that.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .