The work involved decoding neural activity from fMRI scans of the motor cortex. Three regions were scanned, namely the S1, M1 and SMA. These regions were scanned while a participant performed hand movements for the rock-paper-scissors games. The aim of the project is to classify which brain activity pattern represented by voxels is rock, paper, or scissors.
We used 2 linear (Logistic Regression and SVM) and 2 non-linear (KNN and Decision Trees) classifiers and observed that the linear classifiers performed much better than the non-linear classifiers. This is because fMRI data is linear.
The code consists of two notebooks:
- hyperparam_selection.ipynb, which was used to select the best hyperparameters for training the classifiers; and
- project decoding brain activity.ipynb, which is mucher neater, uses the best hyperparameters for the 4 classifiers choosen.