Hi Ajit,
Think about this another way. Your weight calculation is the code span times the grams per code. Grams per code is the long decimal number and is roughly 55.9mg per code (count) which is your error. In other words you cannot resolve to the desired level as there can be a +/- 0.0559 gram variation. The only way to improve this is to reduce the noise. Averaging your results may help if you have eliminated external noise sources. Averaging will extend the measurement time and the number of averages needed will have to be discovered by experimentation. I'm guessing at least 4 points will be needed for the averaged result.
Best regards,
Bob B