Capstone-Brainwave_Controller

Current Benchmark (TRY TO IMPROVE IT!):

• SVM Hyperparameter Value: SEGMENT_SIZE:3, Kernel:'rbf', C:45
• training_accuracy 0.9931437277805993
• test_accuracy: 0.9827411167512691
• test F1 score (each class): [0.9978678 0.98168498 0.97272727 0.97864078]
• test F1 score (weigted): 0.9827235943012146

TODO:

1. Try to improve the overral training accuracy and F1 score.
   • Hyperparamter tuning (SEGMENT_SIZE, model related parameters, etc.)
   • Try other classifiers (Neural Networks(MLP, 1-D CNN, RNN), Decision Tree, etc.)
2. Try to use less features to achieve resonable accuracy (PCA).

• Mapping: stop -> 0, left ->1, right -> 2, forward -> 3. (We do not need "backward" for now because this can be achieved by turn left or right twice and go forward)
• I used Muse 2014 and MacOS for code development.

Setup:

1. Download and install Muse Developer tools
2. Download Python
3. Connect Muse to your laptop using Bluetooth

Gather EEG data:

1. Open one terminal and bridge Muse data to your localhost

muse-io --device Muse --osc osc.udp://127.0.0.1:5000

2. Open a new terminal, activate Python, and run the server script

python server.py

3. Hit Ctrl + C to stop the server; the CSV files should be created under folder Data