Time Series Classification Advice

Hello, I'm fairly new to Machine Learning and I could do with some guidance on how to tackle a problem I'm on.

I have a significant number of time series data which I have measured and I want to develop an algorithm to classify them. A previous project I worked on allowed me to identify certain features in the data, and then use the Classification Learner to generate a model, which accurately predicted which set each time series belonged to.

At the MATLAB expo, I saw that with Deep Learning, I could have an algorithm automatically attempt to identify these features for me. The current data I have is much more complicated, making analysis by myself difficult. Similar to having MATLAB automatically identify the numbers 1 through 9 from a set of images, I want MATLAB to (for example), label a set of time series from 1 through 9.

How do I utilise the tools in the Neural Network toolbox to do this? I'm assuming I want to use the Clustering feature but I'm very overwhelmed by this.

Any advice would be much appreciated.

Thank you

[N/A]

Hello Mr.DK, I have a similar problem as you, in your question you mentioned that u used the classification learner app to identify certian features in your data, could you be able to share that example (personally), that could help me solve my problem hopefully.

Thanks in advance

Bayyu

Answer 1

[Bernhard Suhm]

For deep learning with time series, practitioners are using LSTM. Here's an example in our documentation that shows you how to approach this. You will still need large amounts of labelled data to train, just like for the digit classifier where the MNIST dataset has 6k examples per digit.

[Veronica Taurino]

Hello. I can't see any documentation you talked about. Was it deleted or something?

Thank you in advance