研究背景：女性與青少年是霸凌的高危族群，其復原力高低涉及事後身心適應。為此，本研究調查被霸凌女大學生的復原力、身心健康與腦波之關聯，並輔以機器學習之應用。研究方法：以被霸凌女大學生為對象，依復原力量表分數分為高（前50%）與低復原力組（後50%）。經實驗設計誘發被霸凌經驗，檢驗兩組腦波不對稱分數與情緒狀態差異。同時以腦波不對稱特徵，比較邏輯斯回歸、k-近鄰演算法（KNN）、決策樹與支持向量機（SVM）辨識復原力的準確性。研究結果：高低復原力者各20位。高與低復原力者的額頂葉（C3 vs. C4與 P3 vs.P4）不對稱分數具有差異（all p < .05），創傷壓力程度與感覺運動區（C3 vs.C4）α不對稱分數有關（p = .013）。高復原力比低復原力者自評有較佳身心適應能力（p = .04）。此外，KNN能提供最高的準確率，可達88%。結論：霸凌被害人身心適應預後狀況與復原力高低有關，且涉及神經生理機制（如大腦不對稱活動）。以KNN搭配不對稱腦波可用於霸凌被害人的身心健康狀態追蹤。

Research Motivation and Objective

Bullying has become a major public health concern worldwide. Personal characteristics, such as physical appearance, racial or ethnic identity, gender expression, or family circumstances, can increase individuals’ vulnerability to discrimination, often serving as a starting point for bullying experiences. Notably, young girls are often overlooked as victims of bullying, even though they frequently experience relational and subtle forms of victimization that may be less visible than those encountered by boys. Resilience is widely recognized as a critical psychological capacity that supports individuals in coping with and adapting to stressful or adverse conditions.

Based on previous research, bullying is linked to alterations in both brain structure and function. In clinical settings, electroencephalography (EEG) offers a relatively convenient and low-cost approach for assessing neural activity. Furthermore, biomarkers are increasingly incorporated as predictive features in supervised learning models, a core methodology in machine learning. Accordingly, this study examines the association between resilience levels and atypical EEG activity in individuals with bullying experiences. Furthermore, EEG-derived features were incorporated into supervised learning algorithms to differentiate between high- and low-resilience groups.

Literature Review Bullying and Mental Health

With the widespread use of digital technologies, cyberbullying has further transformed how aggression is expressed and encountered. Unlike traditional face-to-face bullying, cyberbullying occurs through text messaging, social media, and other online platforms, giving perpetrators persistent and unrestricted access to victims regardless of time or location. Gender differences have also been observed, with females particularly vulnerable to relational forms of victimization, including social exclusion, rumor spreading, and manipulation of peer relationships. A robust body of research shows that individuals who experience bullying, whether traditional or online, are at heightened risk for depression, anxiety, low self-esteem, suicidal ideation, and social withdrawal. These psychological and emotional difficulties often disrupt academic functioning, interpersonal relationships, and overall wellbeing. Besides, bullying can lead to alterations in brain structure and function, thereby increasing the risk of subsequent mental health problems. In this context, resilience plays a crucial role in determining how individuals cope with and recover from trauma and stress. Some evidence suggests that females may demonstrate stronger resilience, partly due to their tendency to engage more actively with social and environmental contexts.

Limitations of Conventional Assessments in Bullying Victims

Research consistently indicates that individuals with high resilience adapt more effectively to bullying experiences and exhibit fewer symptoms of depression and anxiety. In contrast, those with low resilience are more vulnerable to developing severe and persistent psychological difficulties. These individual differences are also reflected in neural activity; for example, greater left-frontal activation has been associated with higher levels of resilience and more effective emotional regulation during stressful or traumatic events. However, traditional assessments of resilience rely primarily on self-report measures, which are susceptible to response biases and social desirability effects, potentially limiting their accuracy. Due to these methodological limitations, the psychological distress of bullying victims may not be sensitively or accurately identified.

Applications of Electroencephalographic Technology and Supervised Learning in Mental Health

EEG is a relatively objective measure for evaluating mental states. Research has shown that EEG asymmetry, characterized by high activity in the left hemisphere, is associated with positive emotions and high resilience. Because EEG captures real-time neural activity, it can reveal subtle emotional and cognitive processes that traditional self-report tools often miss, providing a valuable foundation for data-driven mental health assessment.

Research Method

Bullying is known to occur among adolescents, and previous research indicated that females demonstrate higher resilience than males. Accordingly, we recruited 40 female undergraduate participants from a university located in central Taiwan. Participants are divided into high- and low-resilience groups based on their scores on the Adolescent Resilience Scale (ARS), with the upper 50% assigned to the high-resilience group and the lower 50% to the low resilience group. We designed a protocol to record each participant’s EEG activity while they recalled their bullying experiences. The procedure consisted of four stages: baseline, preparation, induction, and recovery. In addition, participants completed the Tung’s Depression Inventory for College Students and the BSRS-5 to assess changes in their emotional state. EEG data were recorded from 8-channel electrodes places according to 10-20 system at F3/F4, C3/C4, P3/P4, and O1/O2. Absolute power was extracted from the delta, theta, alpha, beta, and gamma bands. Brain asymmetry was computed using ln(R) ln(L), with higher values indicating the left-hemisphere has high activity.

Regarding statistical analysis, independent samples t-tests were used to analyze the difference in EEG asymmetry and questionnaire scores between the high- and low-resilience groups. Pearson correlation was performed to examine associations between asymmetry and questionnaire scores. For the supervised learning analysis, EEG asymmetry features were used as inputs to four classification algorithms—decision tree, K-Nearest Neighbors, Support Vector Machine, and logistic regression. The dataset was divided into an 80:20 train–test split using a fixed random seed (random state = 42). Cross-validation procedures were applied, and the Area Under the Curve (AUC) was computed to determine the best-performing algorithm.

Results

Forty female participants were equally assigned to the high- and low-resilience groups, with no significant age differences between them. The results indicated that the experimental procedure successfully re-evoked bullying experiences, as evidenced by significant increases in occipital fast-wave activity during the recall stage (p < .01), suggesting effective induction of visual imagery. Compared with the low-resilience group, high-resilience individuals exhibited greater activation in the left sensorimotor region in the delta and theta bands, as well as stronger left-parietal alpha asymmetry (all p < .05). In addition, sensorimotor alpha asymmetry was positively correlated with emotional intensity on the VAS (p < .05). Machine-learning analyses further showed that EEG asymmetry patterns reliably classified resilience levels, with the K-Nearest Neighbors model achieving an accuracy of 88%.

Discussion and Recommendations

This study is the first to employ EEG data to investigate the neurophysiological underpinnings of resilience and to incorporate these features into supervised learning models for classifying resilience levels in individuals who have experienced bullying. The EEG results demonstrated that individuals with high resilience exhibited greater activation in the left sensorimotor–parietal regions, supporting the notion that resilience may represent a personality related characteristic associated with brain asymmetry. Moreover, the leftward asymmetry observed during the recall of bullying experiences, accompanied by heightened parietal activation, suggests that individuals with different levels of resilience may engage distinct cognitive or emotional strategies when retrieving trauma-related memories. Emotional intensity was also positively correlated with left sensorimotor asymmetry, indicating that high-resilience individuals may transform distress into approach-oriented motivation through more effective emotion regulation strategies, such as cognitive reappraisal. Finally, a machine-learning classifier based on sensorimotor–parietal asymmetry features achieved an accuracy of 88% in distinguishing high- versus low-resilience individuals, suggesting that neural asymmetry may serve as an important biomarker for understanding individual differences in resilience. However, the interpretation of these findings should be approached with caution due to the limited amount of available data.