I would use a single stage butter worth filter on the data I assume you have a known sample time interval, so just google code samples in C for simple butterworth filters utilities that to limit the effects of random results etc. Alternatively use a moving average with dead band on the raw data this however will not exclude random read results that will occur due to extenrnal influences.
So the ideal is a filter first followed by a moving average IMO.
Happy coding //
8) 8)