I have a matrix of samples to classify, samples are matrix columns and features (noisy or estimated features) are matrix rows. On the other hand, I have a lookup table for correspondence between noisless (or theoritical values of) features and classes.
Classification process is very simple: for each sample, corresponding class is that which the features (from lookup table) are the closest to those of the sample.
It seems to be an unsupervised classification, can you confirm ? May it be semi-supervised ?
Best Answer
In supervised learning, you try to predict a label. As long as you do it, even if you don't actually learn, this is your framework.
In semi supervised learning, you have a small set of labeled data and a much bigger unlabelled dataset you can use in order to improve you predication.
In unsupervised learning, there are no labels and therefore you don't try to predict. The goal is usually to infer a structure of the data. There is no such a thing as unsupervised classification.
Since you try to predict labels, you are in the supervised learning labels.
Sometime, the use of lookup table are considered to be an expert system or usage of domain knowledge. At very least, these terms make it sound better.
Since you wrote that you use "closest to those of the sample" I deduce that you have a distance function. Working in such a way is close to 1 nearest neighbor which is definitely a supervised learning algorithm.