This is not a MATLAB question. MATLAB can help you implement after you have decided on a model, but it cannot find the model for you. (The closest would be the System Identification Toolbox)
Mathematically, given any finite list of finite precision measurements, there are a literal infinite number of functions that can fit the equation to within round-off error. (In the past I have posted a constructive proof of this.) And because there are an infinite number of possibilities, the probability that any particular version is the "right" equation... is ZERO.
Therefore you can never start with the data and use it to generate equations until one matches the data. You always need to instead start with knowledge of the physical system. If you know the general equation for how viscosity changes with temperature but it has some unknown parameters, then you can fit the parameters to the data in a meaningful way.
The alternative is to use a neural network to start from the data and approximate the function. The result can sometimes be analyzed to figure out a meaningful equation... but certainly not always (not even most of the time.)
