Finding optimal parameter for data smoothing

Hi, I'm trying to find the optimal window length for smoothing noisy GPS data (x,y coordinates).

I'm using Savitky Golay Filter, and I have to set the parameter 'window length'.

On each step, I used different window sizes for smoothing noisy data, and calculated Mean Square Error as below. (increasing 1 window size for each step)

MSE is calculated between smoothed data and noisy data.

As the window length increases, the MSE also increases constantly.

And there is a elbow/knee point marked as '*'.

In this case, can I set the elbow/knee point of the MSE graph as 'optimal window length' of the smoothing filter?

What does it mean to set knee point as an optimal window length in this case?

Many thanks!

[Image Analyst]

As the curve gets smoothed more and more, the curve will get flatter and flatter, and the elbow point will move. How are you finding hte elbow? And how will you know if it's "optimal"?

[Mathieu NOE]

it's probably not "optimal" as it simply says that beyong that point the smoothed curve is getting way too flat so the error is even accelerating. The optimal point is somewhere between the start and this point but it may be very difficult without seeing the raw data what can be called "optimal" as different persons may have different opinions of how much the data need smoothing.

other alternatives could be spline / polynomial fiiting ... ?

have raw data to show ?