Time Series Classification with a convolutional neural network ?

Hello Community,

my lack of experience in machine learning leads me to ask you guys. I want to classify different time series.

A bit background:

I measured the movement of my skateboard while doing tricks with an IMU ( gyroscope, accelerometer, magnetometer). I did 4 different tricks, each 50 times. Than i cute the long time series with all tricks in it into samples ( 1 sample is 1 trick each with the same lenght) and sorted by trick (class).

My goal would be, that I can show my machine data and it can classify by its class ( type of trick).

I read alot about this things and I think there are many ways to acchive this. I read that a convolutional neural network or a decision tree could be a good solution. What do you think - any suggestions ?

Would appreaciate all answers.

Have a good day!

Answer 1

[Aditya Patil]

As the data is temporal, you can use one of the sequence classification models. For example, you can use LSTMs (Long Short-Term Memory Networks). See the sequence classification using Deep Learning example https://www.mathworks.com/help/deeplearning/ug/classify-sequence-data-using-lstm-networks.html.

Alternately, if you know that the data can be represented well in structural format, you can use any of the classification algorithms/models available in Statistics and Machine Learning Toolbox, or in Deep Learning Toolbox.

I also recommend looking for pretrained models for this task and trying transfer learning.