Confusion Matrix showing network accuracy in classifying left and right hand movement

Confusion Matrix showing network accuracy in classifying left and right hand movement

Artificial neural network as a method to classify hand movement from EEG signals

Brain-Machine Interface (BMI) technology has the potential to restore physical movement but it requires an accurate classification of human brain patterns in real time to generate commands for digital devices. This work details the accuracy of Artificial Neural Networks (ANN) in classifying motor imagery signals recorded from Electroencephalography (EEG) signals at a 256 sampling rate. The recorded data was processed through signal processing techniques and features were extracted for use as feature vectors in the network. The results of this study show that ANN’s given the right training algorithm and a number of hidden neurons, can adequately classify EEG signals. 

The Neural Network created is a three-layer feedforward network with a sigmoid (tansig) function in the first hidden layer, logsig function in the second hidden layer and softmax in the output layer

To train the signals, I used a scaled conjugate gradient algorithm (Trainscg) and also a resilient backpropagation (Trainrp) to observe differences in the algorithm chosen. I also used different number of neurons in the hidden layers to compare accuracies. These are broken down further in the table below

 

ann1.png
Artificial Neural Network structure.

Artificial Neural Network structure.

Table showing the breakdown of different number of neurons in the hidden layer using the two training algorithm and their accuracies

Table showing the breakdown of different number of neurons in the hidden layer using the two training algorithm and their accuracies