Classification approach ideas for error detection

Hi,

I want to make "something" for data classification, for an error-detection mechanism. I have a set of input data (different error simulations, eg. heating a part of a machine and taking sensor data), eg. err1...err4. Each beeing a set of features (sensor data, 20 different points) across multiple measurment points, so a 20x400 vector for example. For each of those i have a set of output data, 3 sensors logging data, so 3x400. The sensors for output data are different than those for input.

The idea is that later, I want to heat one component up and see in those 3 sensors of the output, that I had heating of a component and not something else.

I'm kind of new to this and I still don't have a clear direction to follow.

I need something to be able to input all of the data or even 4 different classifications, and to tell me, ideally, the percentage of being in one of the 4 error mechanism.

So far I've read into Fuzzy classification and the Fuzzy Toolbox (there I have problems with entering this data, I don't think it is possible for this usecase). Next the MATLAB classification learner, but there I have only one response? Now I'm reading into SVMs... Any ideas on what I should read into, check out or try? I don't want to use neural networks, but other machiene learning should be fine.

Thank you.