I want to train a classifier that is able to identify different activities based on EMG data that is sampled at 41Hz with 8 channels, I read this paper that summarizes a lot of different approaches to clean the data, transfer the data and extract features as well as classification algorithms that work well with emg data.

After carefully thinking about my approach here is how I think I should proceed. I want to have a sanity check to see if there are any better approaches or any advice on my current approach.

First use a Daubechies Wavelet Transformation (db2 and decomposition level 4) the data, this method was summarized as best performance based on the paper.

Then I want to extract Time-frequency representations using wavelet packet transform and use PCA for dimensionality reduction.

Finally feed this into either an Adaptive neuro fuzzy inference system(ANFIS) or into an MLP. I'm not completely sure which one to use yet.

Is there anything you would suggest to change or improve on with this method? Any advice would be great. Thank you.

  • $\begingroup$ sanity check: "best in the paper": does the metric the paper uses for "goodness" coincide with your own purpose? $\endgroup$ – Marcus Müller Feb 9 '20 at 9:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.