Smoothing algorithm (for data from cows' ears' sensors)

[darren0123]

Hi ?
I'd appreciate your help.
I have a bunch of livestock with an ear sensor on each, that sends me data about how much heat is going out of the ear.

I wanna take this data from each livestock and smooth it.

My question is, how to know which smoothing algorithm I should use to get an authentic reliable smoothed data?

Thank you.

 

[mmm4444bot]

Hello. It may be helpful for tutors to see a sample of that data. ?
[imath]\;[/imath]

 

[Mr. Bland]

You may find successes using spline interpolation.

Here's an example from SAS:

 

[BigBeachBanana]

Hi ?
I'd appreciate your help.
I have a bunch of livestock with an ear sensor on each, that sends me data about how much heat is going out of the ear.

I wanna take this data from each livestock and smooth it.

My question is, how to know which smoothing algorithm I should use to get an authentic reliable smoothed data?

Thank you.

Other common methods are simple moving averages (long-term trends) and exponential smoothing (short-term trends).
Depending on your purposes, one method is preferred over the other.

 

[blamocur]

You may find successes using spline interpolation.

Here's an example from SAS:

View attachment 34769

I disagree: interpolating noisy data is rarely meaningful. And I am assuming that the data is noisy otherwise there would not be much need for smoothing.

 

[blamocur]

Other common methods are simple moving averages (long-term trends) and exponential smoothing (short-term trends).
Depending on your purposes, one method is preferred over the other.

To expand on @BigBeachBanana's post: there are lots of ways to smooth noisy data. In addition to simple moving average there are weighted moving averages, which signal processing folks call low-pass FIR filters. Choosing coefficients for such filters is a non-trivial task ("filter design" in the signal processing lingo).
P.S. I am not sure I agree with long/short-term characterization: exponential filters' outputs theoreticalljy depend on all previous inputs (they call them Infinite Response Filters in signal processing).