You should find a mathematical model a model to each of your lines using curve fitting toolbox. Type cftool in the command window to open the app. Generally, you first find a good model visually, and then validate it. Please read the post processing documentation: https://www.mathworks.com/help/curvefit/fit-postprocessing.html
I have fitted the first variable (SOH{1} vs. cycle{1}) and it seems that an exponential function of the second order can be a suitable fit, since the fitted curve is very close to the data points and I have got tight confidence bounds for each parameter.
The histogram plot of residuals also show that the follow a symmetric Normal distribution, which is another sign of a good fit.
After you achieved a good model for each of your lines, you can use fit functions to evaluate it for points outside the data points, i.e., extrapolation.
There is a caveat with extrapolation, that your actual system might have a different behavior in regions that have not been sampled. Extrapolation assumes that your system behaves similarly inside and outside of the sampled data.
