Machine learning-based methodology for multivariate analysis of electroencephalogram

다국어 초록 (Multilingual Abstract)

Analyzing electroencephalogram (EEG) data within an experimental design pose challenges due to the significant variability in complex EEG features. This variability can be influenced by multiple factors, including the experimental design, individual differences among participants, and the presence of artifacts. Machine learning application is increasingly employed to analyze complex EEG patterns by integrating multiple EEG features. As a result, it is crucial to evaluate and compare the effectiveness of the machine learning approaches in analyzing EEG patterns.
The two experiments were designed to explore EEG interpretation based on machine learning application. In the first experiment, the Support Vector Machine (SVM) algorithm was used to investigate whether the effects of brain stimulation would vary depending on circadian rhythm and chronotype using EEG features. In the second experiment, EEG features was applied to three different machine learning algorithms: SVM, Logistic Regressor (LR), and Extreme Gradient Boost (XGBoost) to classify emotional mental states induced by virtual reality (VR) content. Classification accuracy and feature importance were used to evaluate the performance of the machine learning algorithms. Classification accuracy measures the accuracy of the classification results, while feature importance identifies which EEG features are most important for the classification conditions. The findings of two experiments demonstrated that machine learning application based on experimental EEG data was successful with superior performance. Feature importance analysis can help identify the most informative features for classification and improve the accuracy of the model. The results of feature importance in the first experiment found distinct effects of brain stimulation on both circadian rhythm and chronotype, suggesting that these factors play a crucial role in shaping the impact of brain stimulation. The second experiment yielded significant results in terms of feature importance, revealing distinct outcomes that underscored the effectiveness of VR in eliciting specific emotional responses. These findings provide compelling evidence for the powerful impact of VR in evoking targeted emotional states. The observed variations in the effects of VR on emotional responses further emphasize the potential of VR as a valuable tool for emotion induction. Machine learning can be used to identify the neural mechanisms underlying the effects of brain stimulation. In the second experiment, the feature importance analysis showed that different emotional states could be successfully induced using VR contents, suggesting that EEG-based machine learning can be used to study the neural correlates of emotions and their modulation.
The findings in the study demonstrated the potential of machine learning application for EEG data analysis, particularly for classifying and predicting brain states. The study suggested that the selection of appropriate machine learning algorithms and feature sets is critical for achieving high accuracy in classification tasks. The study has important implications for the field of EEG data analysis and machine learning application. By successfully demonstrating the potential of machine learning techniques to analyze EEG signals and identify important features for understanding brain function, the study provides a promising method for future research in this area. Moreover, it can extend to other research fields such as medical or neurocognitive research, beyond EEG data analysis, to provide a prospective method for researching a variety of brain functions.

국문 초록 (Abstract)

실험 설계에 기반한 뇌파 분석은 다양한 요인에 의해 영향을 받아 어려움을 가지고 있습니다. 실험 설계, 개인 간 차이, 잡파 등 다양한 요소들이 뇌파 분석에 영향을 미치는데, 이는 정확한 뇌파분석의 어려움을 증가시킵니다. 다양한 뇌파 특징을 통합하여 복잡한 뇌파 패턴을 정확히 식별하기 위해 기계 학습 접근 방법이 점차적으로 연구되고 있습니다. 뇌파 신호 분석 연구는 기계 학습 기법을 적용하는 데에 있어서도 점점 다양한 접근 방식이 시도되고 있으며, 이에 따라 뇌파 패턴에 대한 기계 학습 응용을 평가하고 비교하는 것이 필수적입니다. 이는 뇌파 신호 분석의 정확성과 신뢰성을 높일 수 있을 것으로 기대됩니다. 본 연구는 뇌파 분석을 위한 기계학습 응용방법을 탐색하기 위해 두 가지 실험을 설계했습니다. 첫 번째 실험에서는 Support Vector Machine (SVM) 알고리즘을 사용하여 뇌 자극의 효과가 일주기 리듬과 일주기 유형 따라 다른지를 검증했습니다. 두 번째 실험에서는 뇌파 특징을 세 가지 다른 기계 학습 알고리즘인 SVM, Logistic Regressor (LR), 그리고 Extreme Gradient Boost (XGBoost)에 적용하여 가상 현실 (Virtual Reality: VR) 콘텐츠에 의해 유발된 감정적인 정신 상태를 분류 가능한 지를 검증했습니다. 분류 정확도와 특징 중요도는 기계 학습 알고리즘의 성능을 평가하는 데 사용되었습니다. 분류 정확도는 분류 결과의 정확성을 측정하는 반면, 특징 중요도는 분류 조건에 가장 중요한 뇌파 특징을 식별합니다. 두 실험의 결과는 실험 기반의 뇌파 분석을 위한 기계학습이 우수한 성능으로 성공적이었음을 보여주었습니다. 특징 중요도 분석은 분류에 가장 유용한 특징을 식별하고 모델의 정확성을 향상시킬 수 있습니다. 특징 중요도 분석은 뇌 자극의 효과가 일주기 리듬과 일주기 유형에 의해 설명될 수 있다는 것을 보여주었습니다. 두 번째 실험의 특징 중요도 분석을 통해 VR 콘텐츠를 사용하여 다른 감정 상태를 유도할 수 있음을 보여주었습니다. 이는 뇌파를 기반 기계 학습이 정서 상태와 정서조절의 신경 연관성을 연구하는 데 사용될 수 있음을 시사합니다. 본 연구는 뇌파 데이터 분석 및 기계 학습 응용 분야에 중요한 함의를 가지고 있습니다. 기계 학습 기법을 사용하여 뇌파 신호를 분석하고 뇌 기능을 이해하는 데 중요한 특징을 식별하는 잠재력을 성공적으로 입증함으로써, 본 연구는 뇌파를 사용한 뇌기능 분석 방법을 위한 유망한 방법을 제시했습니다. 뿐만 아니라, 이는 뇌파 분석을 넘어 의료 및 신경인지 연구와 같은 연구 분야에 확장될 수 있는 다양한 뇌 기능 연구 방법을 제시함으로써 의의를 갖습니다.

목차 (Table of Contents)

• CHAPTER 1
• An Overview of EEG-Based Brain Research and Machine Learning
• Techniques for Investigating Brain Function in the Study ..................... 1
• 1. Introduction .................................................................................. 2
• 1.1 Brain research based on Electroencephalogram (EEG) ........ 2
• 1.2 Machine learning application based on EEG data ................. 6
• 1.3 The goal of the study ........................................................... 11
• CHAPTER 2
• Research Background, Experimental Procedure, Finding, Discussion
• and Conclusion .................................................................................... 13
• 2.1 Experiment 1 ............................................................................ 14
• A machine learning approach
• to find the effects of tDCS(transcranial Direct Current Stimulation)
• by circadian rhythm and chronotype
• 2.1.1 Background of the study ..................................................... 14
• 2.1.2 Method ................................................................................ 19
• 2.1.2.1 Participants .................................................................... 19
• 2.1.2.2 Participant Classification ............................................... 19
• 2.1.2.3 Experimental procedure ................................................ 21
• 2.1.2.4 EEG device .................................................................... 22
• 2.1.2.5 Transcranial Direct Current Stimulation (tDCS) ............ 23
• 2.1.2.6 EEG preprocessing ........................................................ 25
• 2.1.2.7 EEG feature extraction .................................................. 25
• 2.1.2.8 Machine learning application ........................................ 30
• 2.1.2.9 Machine learning performance and feature importance . 31
• 2.1.3 Results ................................................................................. 33
• 2.1.3.1 Machine learning performance ...................................... 33
• 2.1.3.2 Feature importance ........................................................ 35
• 2.1.4 Discussion on the first experiment ....................................... 40
• 2.2 Experiment 2 ............................................................................ 47
• An appropriate machine learning algorithm
• for electroencephalogram (EEG) patterns of different mental states
• induced by VR content (High arousal/Low arousal/Social anxiety)
• 2.2.1 Background of the study ..................................................... 47
• 2.2.2 Method ................................................................................ 51
• 2.2.2.1 Participants ..................................................................... 51
• 2.2.2.2 Experimental procedure ................................................. 51
• 2.2.2.3 EEG device and EEG preprocessing .............................. 53
• 2.2.2.4 EEG Feature extraction and selection ........................... 55
• 2.2.2.5 Machine learning application ......................................... 56
• 2.2.2.6 Machine learning performance and feature importance . 58
• 2.2.3 Results ................................................................................. 60
• 2.2.3.1 EEG Feature selection ................................................... 60
• 2.2.3.2 Machine learning performance ...................................... 63
• 2.2.3.3 Feature importance ........................................................ 66
• 2.2.4 Discussion on the second experiment .................................. 68
• 2.3 General discussion based on two experiments ............................ 75
• 2.4 Conclusion ................................................................................. 78
• References ............................................................................................ 81