Although picture may be worth a thousand words, having your data to work with is worth a thousand pictures.
I would begin with a Fourier transform of your data (the fft (link) function) to determine what part of your data are your signal and what part are noise. Once you have the frequency information, and you know you can filter out the high-frequency noise, a prototype lowpass filter design for you to experiment with is:
Fs = 1000;
Fn = Fs/2;
Wp = 100/Fn;
Ws = 105/Fn;
Rp = 1;
Rs = 150;
[n,Ws] = cheb2ord(Wp,Ws,Rp,Rs);
[z,p,k] = cheby2(n,Rs,Ws);
[soslp,glp] = zp2sos(z,p,k);
figure(3)
freqz(soslp, 2^16, Fs)
filtered_signal = filtfilt(soslp, glp, original_signal);
This design has a much steeper transition region than your Butterworth design, so it may have less effect on the part of your signal you want to retain. Make the appropriate changes in the parameters for your signal.
Best Answer