Classification Model for EEG Signals

Supervised Machine Learning Model to Classify Brain Waves

In this project we used 10 EEG (Electroencephalogram) Sensors non-invasively surrounding a subject's head. We also used IMU (Inertia Measuring Unit) attached to the subjects wrist to aid labeling of data.

For reading data, we place a Reference Sensor in the middle. Equidistant from all sides of the subjects head. We also set a ground sensor on the left ear. All the sensors are connected to a hub which is connected to a computer that can record data. An accompanying IMU on the users wrist is connected to the computer recording data.

Subject is instructed to think for a set amount of time and then write for a set amount of time. And the data is collected from these sensors for multiple trials.

​

After EEG training data is collected and labelled we pass it through a Bandpass Filter at 1Hz and 45Hz. We use the accompanying IMU data to label the data and only take the first 2 seconds of EEG data after every changeover. We then implement a CSP algorithm (Common Spatial Pattern) to calculate the CSP weight of the of our training dataset. We use the CSP weights to plot a topological plot. 

​

For this project we used several algorithms to find the one that provides the best accuracy. We used variance of CSP filtered data with Fisher's Linear Discriminant Alalysis (LDA) as one classifier model. We used just LDA as another classifier model, And we used SVM (Support Vector Machines) using 2nd order and 6th Order Polynomials as two other classifying models. We train all 4 models with the training data.

​

We then collected some test data from the same subject and labelled the data. We then put the data through our models and compared the model output to the labeling to find the model best accuracy which ended up to be 2nd order Support Vector Machine