Discipline: Technology and Engineering
Subcategory: Biomedical Engineering
Session: 3
Room: Exhibit Hall A
Abigail Turcheck - Arizona State University
Co-Author(s): Jesus Cruz-Garza, University of Houston, Houston, TX; Dr. Jose Luis Contreras-Vidal, University of Houston, Houston, TX
The observation and analysis of human creativity are difficult in a lab setting due to the inhibitory nature of experimental controls and devices that record brain activity. This experiment aimed to mitigate this problem by using a mobile brain-body imaging (MoBI) dry-electrode EEG cap to record an artist’s activity during her 18-month-long art project. After preprocessing of the data and artifact removal, a machine learning data analysis process was applied to the total dataset [1]. This data analysis method aims to correctly classify the various activities the artist was performing over the span of the dataset, given the EEG data collected from the channels of the dry-electrode mobile EEG cap. Due to the large amount of data collected, a process involving reiterative data resampling was used to make the process more time-efficient while maintaining accuracy. Precautionary measures to avoid data leakage were implemented by separating data files from different days so that the testing set contained data from different days than the training/validation set. After obtaining a cubic SVM classifier from a random sample of 300 data points from the training/validation set, average accuracy values for validation and testing were determined by repeating classification with 50 iterations of random resampling. Additionally, feature selection was performed on randomly resampled data using a maximum relevance-minimum redundancy (mRMR) algorithm, determining the most relevant EEG channels and frequency bands to the classifier. 26 channels were found to be the most relevant. The 26 features had classification accuracy of .4545 with standard deviation of .0464. It was found that channels in the right temporal lobe, right parietal lobe, right frontal lobe, and the left parietal lobe had the most relevant features to the classification of activities. All frequency bands were relatively equal in their relevance. This process of data analysis demonstrates the ability of machine learning to successfully classify creative activities solely using EEG data. Further studies using MoBI EEG with new subjects are needed to assess the origins of authentic creativity in the human brain to determine if the selected relevant features vary significantly throughout the population. [1] Cruz-Garza, J. et al (2017). Deployment of mobile EEG technology in an art museum setting: evaluation of signal quality and usability. Frontiers in human neuroscience, 11, 527.
Funder Acknowledgement(s): This project was funded by NSF Award #1757949 to the University of Houston Cullen College of Engineering (Houston, TX, USA) through the 2019 REU in Neurotechnology Program.
Faculty Advisor: Dr. Jose Luis Contreras-Vidal, jlcontr2@central.uh.edus
Role: I completed feature extraction and selection on the dataset, helped design the reiterative resampling data analysis pipeline with my mentor, wrote MATLAB programs to run the data analysis pipeline, and created/compiled figures from the results of the machine learning and resampling processes. I then interpreted these results in regards to the functions of different areas of the brain.