- 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:
- 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.)
- 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.
- Download and install Muse Developer tools
- Download Python
- Connect Muse to your laptop using Bluetooth
- Open one terminal and bridge Muse data to your localhost
muse-io --device Muse --osc osc.udp://127.0.0.1:5000
- Open a new terminal, activate Python, and run the server script
python server.py
- Hit
Ctrl + C
to stop the server; the CSV files should be created under folderData