Acceleration Integration always shows same error

[AnaRhisT94]

When running acc.'s integration script it always shows an error around 0.8m displacement in the y-axis, doesn't matter if I move it or not.
I commented mag's values because it makes rpi zero to work really slow and not sample fast enough, which results in error in the signal filtering later too.

I also load the calibration coeff's after the calibration is done.

the output:

Data Acquisition Stopped
1540 - number of samples
5 - seconds
308.0 - division (1540/5)
Sample Rate: 308Hz
Displace in y-dir: 0.90m

[AnaRhisT94]

After finding the bias I've reduced the error to 0.1m after 5 seconds of stationary.
Any interesting ideas how can I further reduce the error to 0.01m~ like shown in your GIF example?