classification based on evolution of features in time I have signals, from this signal I do windows analysis and extract features. So from 1D signals of 882000 points I get 200 vectors (200 windows of analysis) of 3 values (200*3 matrix, Let's call "InputMatrix").
I would like to make classification (in an objective of machine learning, with supervised learning), to separate 6 types (6 classes) of signals.
The differences are the time evolution of my extracted features.
What would be a good methods for this ? SVM ? Artificial neural network ?
My main question is about taking in account time evolution
2nd question : what if I get different size of "inputMatrix" for each sample ? 
 A: If I understand correctly you have 200 samples described by 3 features.
This in fact represent only one signal to classify into one of 6 categories. I then suppose you have many 1D signals to classify, which represent your dataset.
You think that your signal category can be inferred from the describing features evolutions. Even if you don't think it's smart to consider the full matrix as an input, let me try to change your opinion.
This 200 x 3 represent one sample, let's think about this matrix as an image (with your feature accordingly normalized). It could be interesting to experiment a solution with an mindset of image classification, in the sense that you would be using CNNs. By using 1d convolution of TBD size n (so with a filter of size n by 3 for first convolution layer) you could treat individual temporal slices as discriminant classifying patterns. You would be effectively calculating theses different 'temporal' activations via convolutions, merging them in your architecture and finally applying a label to it. The only difference with a classical image classification via CNNs would be that it has no sense to explore your feature dimension (so your convolution hase only one degree of freedom), only the temporal is here relevant.
This method of 1d convolution with CNNs has been used in music genre classification with great success. Your 200x3 is your music, your 6 classes the genres.
