EEG-Based Human Recognition During Video Game Viewing

Abstract

This project aims to predict the identity of individuals based on EEG data recorded while watching video game footage. For this purpose, the dataset consists of EEG recordings from 31 participants while they watched five different video games.

Data Description

The data contains raw EEG signals from electrodes placed on the scalp, individuals' head movements, the power of the raw EEG signals, participant IDs, and game names. Data Availability Statement: The data is confidential.

Methods

LSTM Deep Learning Models

In this step, the empty Motion and Power columns were filled with the previous values, and the game names were adapted to the algorithm using one-hot encoding. Then, an LSTM model was used to distinguish between individuals

Why LSTM: LSTM is an algorithm used for time series data. Compared to RNNs, it can remember longer sequences of data due to its forget gate mechanism. This capability makes LSTM a better choice for the task.

Traditional Machine Learning Algorithms

We will try to achieve success with traditional ML models by performing our own feature extraction on the data we provided to the LSTM model.

In time series data, we will use the sliding window technique to extract certain statistical features to transfer information from previous data to the ML model. For example, features such as RMS, min, max, mean, standard deviation, kurtosis, and peak-to-peak will be extracted.

Choosing the window size is important here. Values like 100 and 200, which are multiples of each other, have been selected according to the data. There is overlap in the windows.

Conclusions

In this study, we evaluated the performance of various models for human recognition based on EEG data collected during video game viewing. Both deep learning and traditional machine learning models were explored. Specifically, we tested Decision Tree, Random Forest Classifier, and Logistic Regression models. Our results indicated that both deep learning and traditional ML models were successful in achieving the desired outcomes.