Abstracts—Brain-computer interfaces have been investigated for more than 20 years and have great potential to develop applications for physicians to diagnose diseases or patients with severe neurologic disabilities to return to interact with society. To gain those purposes requires technics to analyze the EEG data as well as an algorithm to train the model for identifying the patterns or controlling the devices. TensorFlow is a machine learning developed by Google team for internal use and was released for public use in 2015. Since it can train and test on deep learning neural network, it can be used for EEG data. This project used TF-Keras and TensorFlow-DNN to train the models for classifying brain states using EEG data. Neurosky Mindwave Mobile headset and a new device developed from Micro:bit were the recorders for EEG signals in the project. Several technics such as min-max normalization, Ensemble Empirical Mode Decomposition (EEMD), extraction were applied to analyze the recorded EEG data. The results show that the accuracies of TensorFlow-Keras and TensorFlow-DNN models are 97% while the results from XGBoost is 98% when classifying the EEG data from Micro:bit device. The result confirms the ability of application of TensorFlow in identifying EEG data. The technics for processing data contributed to the above results are min-max normalization and data extraction. Moreover, we also verify that the low-frequency drifts in the recorded data is essential to identify the brain states using EEG data. The results also show the application of IMFs generated from EEMD technic as features to build the models for classifying brain states using the EEG data.
 

Index Terms—TensorFlow, EEG, XGBoost, TensorFlow-Keras (TF-Keras), TensorFlow-DNN (TF-DNN), Ensemble Empirical Mode Decomposition (EEMD), Neurosky, Micro:bit, Brain-C omputer Interface (BC I)

Cite: Phuong Huynh, Gregory Warner, and Hong Lin, "Effects of EMD and Feature Extraction on EEG Analysis," Journal of Advances in Information Technology, Vol. 11, No. 1, pp. 26-34, February 2020. doi: 10.12720/jait.11.1.26-34

Copyright © 2020 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.