the simplest method is to simply remove the mean value of your signal.
another approach is to use a high pass filter, this can also be useful if you want to remove very low frequency drifts or motion effects .
tested on you x data , you can easily copy paste on y and z data;
t_ac = (0:samples-1)'*dt;
[B,A] = butter(N,2*fc/fs,'high');
title('acceleration (unit ?)');
ylabel('Amplitude (unit ?)')
legend('raw','mean removed','high pass filtered');