Use the statistical toolbox

Can directly realize 1 multivariate linear fitting regression function, very simple. This method is similar to the linear least square fitting of data, and is actually related to the overdetermined equation. But in practical problems, most of them can't be described by linear relationship.

2 Establish your own functional relationship, use nonlinear fitting function nlinfit, and other functions such as fittype lsqcurvefit, but you have to figure out your own functional expressions.

I usually use artificial neural network toolbox to deal with it. By adjusting the parameters of BP network, satisfactory results are obtained. But this method produces a mathematical model, not a specific function expression.