This repository contains the code for selecting the best time-domain EMG features using the novel and some classical feature selection algorithms.
The dataset has been taken from the publicly available NinaPro DB2 database in the following link: https://ninaweb.hevs.ch/.
The PR model involves three major stages like signal preprocessing (e.g., denoising), feature engineering (feature extraction & selection), and signal classification based on Machine/Deep Learning (ML/DL) algorithms.
Our work is focused on selecting the best EMG-based time-domain features using the novel Neighborhood Component Analysis-based feature selection (NCA-FS) method proposed in our conference paper. For the details, you can access the paper and observe our results.
As an extension, the NCA-FS method have been compared with other classical filter feature selection methods like Relief and mRMR and showed superior classification performance for a decision tree-based classifier for several scenarios.