Tetris-EEG

This repository contains code to run an adaptation of Tetris that dynamically changes the game speed based on Machine Learning predictions of the user's current Arousal and Valence every 4 seconds. This project was originally built to run an experiment called EEGoFeedback: Dynamic Flow Maintenance through Real-time EEG Emotion Prediction.

Repo Structure:

.
├── dreamer_model
|   ├── datasets  # all generated csv go here
|   ├── ML  # contains models, utils, and training split functions
|   ├── features.ipynb  # generates features tables from DREAMER
|   ├── predictor_model.ipynb   # train model to predict in-game
|   └── run_models.ipynb    # notebook to compile all models
├── experiment
|   ├── EpocX   # contains code to get and manipulate data from EEG
|   ├── experiment.py   # functions for syncing EEG data with game
└── tetris
    └── main.py

Initial Setup

Once you have set up your python environment, run

pip install -r requirements.txt

Playing the game/Data Collection

Current tetris implementation only tested with Python 3.13.7

To play the game, set the current working directory to Tetris-EEG/tetris, then run main.py. Make sure that an Emotiv EPOC X headset is close by and powered on.

python -m tetris

Controls

Button	Action	Description
Left Arrow	Move Left	Moves the current piece to the left
Right Arrow	Move Right	Moves the current piece to the right
Up Arrow	Rotate Piece	Rotates the current piece
Down Arrow	Soft Drop	Increases fall speed of the current piece
Space	Hard Drop	Instantly lands and locks current piece
C	Hold/Swap	Stashes current piece in the Hold cell, swaps with piece in hold cell if available

EEG Prediction Module

This repository also contains a dedicated subproject for EEG-based emotion recognition in the directory EEG Prediction.

Inside, you'll find:

• notebooks/ → experiments with classical ML and deep learning models (MLP, MIL, LSTM, ANN/SVM/RF/KNN)
  • Feature Importance Analysis → comprehensive analysis to identify which EEG features most impact valence/arousal prediction
  • Filtered Dataset Creation → generate optimized datasets with only the most important features
• scripts/ → reusable Python scripts for baselines and dataset utilities
• datasets/ → expected DEAP dataset structure (not included, must be downloaded separately)
  • filtered_features/ → reduced feature sets for faster training and better interpretability
• requirements.txt → dependencies for running notebooks and scripts

Key Features

• Multiple ML/DL approaches for emotion recognition (valence & arousal)
• Support for both DEAP and DREAMER datasets
• Comprehensive feature importance analysis using 5 different methods
• Pre-filtered datasets with top N important features (configurable)
• Visualizations for feature rankings, frequency bands, and brain regions
• Automatic merging of DREAMER feature files (features_table + IMF features)

Please read the EEG Prediction README before running experiments or submitting changes.