Classification Model for EMG Signals

Supervised Machine Learning Model to Classify Muscle Signals

8 Pairs of EMG (Electromyography ) Sensors are placed close to the Elbow of test subject to record muscle activity.

Electromyography Sensors work in pairs, a reference node and a non-reference node. And the potential difference between the two nodes are the signal read.

These sensors are connected to a hub which is connected to an Arduino MCU to record data at a sampling rate of 1KHz for 20 seconds.

For this purpose we are classifying the movements into 5 classes: Neutral, Wrist Extension, Wrist Flexion, Radial Deviation and Ulnar Deviation. 

After getting the data we have 8 streams of data at 1KHz. For better prediction and training the model and  we add more steams of data. A data stream with RMS (Root Mean Square) of the data with a moving window of 100ms is added. Along with a data stream of Waveform length using a moving window of 100ms. We also add AR ( Auto-Regressive) coefficients as a data stream.  We then label the data to train our LDA Model ( Fisher`s Linear Discriminant)

​

On the same person we then collected 5 seconds of testing data. We then fed the data to our Model and had a best case scenario efficiency of 67.1%