You can do something like this:
sfcn = @(t,pars) pars(1)+pars(2)*sin(pars(3)*t+pars(end)); % sine-like function with offset and phase-shift
fitfcn = @(pars,y,t,sfcn) sum(( sfcn(t,pars) - y).^2); % Sum-of-square function
Clipped3 = clipped(:,3); % Your thirs column
t = 1:numel(Clipped3); % Some time-array for uniformly sampled data
% Identify your clipped data points:
iBad = find(Clipped3>290|Clipped3<2);
Clipped3(iBad) = []; % Cut those points out
t(iBad) = []; % from data and time-array
pars0 = [150,500,2*pi/(46.7),pi/2];
% this fitting is painfully sensitive to bad start-guess so the above line
% we set the average at 150, the oscillating amplitude to 500, and the angular
% frequency to fit with a period-time of 46-47 time-steps, and the phase-shift
% to pi/2 since your data starts at a peak not a zero
% Fit by minimizing sum of squared residuals:
parsOK = fminsearch(@(pars) fitfcn(pars,Clipped3,t',sfcn),pars0)
%parsOK =
% 166.1 240.25 0.13416 1.7285
plot(clipped(:,3),'b-') % initial curve
hold on
plot(t,Clipped3,'b.') % unclipped points
% Residuals:
plot(t,Clipped3'-sfcn(t,parsOK),'r.')
% Well-fitting sinusoidal curve
plot(t(1):t(end),sfcn(t(1):t(end),parsOK),'r')
HTH
