Hello,
I have a set of 12000 1-D signals (time series). They are measured under two conditions (Condition 1 and Condition 2, see figure).
I would like to use machine learning methods to automatically distinguish my data set into one of these two conditions.
I would like to create the detector which does not need to establishing features, but takes directly the samples of the signal as an input, similarly to image recognition in convolutional neural networks (CNN).
Is it possible to use CNN in my problem? Or is it a different method?
From what I understand , your 1D signal can be taken as 2D (x vs y-axis) and you can use CNN .
Using condition-1 you can train the network and then using condition-2 you can compare the feature set for predictions .
Hi,
I wouldn't like to analyze these signals as images. I thought rather about time series classification which will be implemented into software of mobile device measured this kind of signals. I think the conversion 1-D data to 2-D images would be inappropriate and costly. But maybe I'm wrong... I'm not so familiar in this subject.
Mr Hilal, could you tell something more about " kodified- structre CNN "?
I have used CNN with images (2D) although you can set up the input there (I am not confident about 1D) . Yes it requires 1000s of images .
Next, with ANN I am sure you can use 1D . Variants of it too. Here data can be of any length . Again , you should have training input and target samples . Then a subset for prediction.
NN is basically a regression (mathematically) .
- Badges
- Science topic