전체메뉴

Journal of Pharmacopuncture

Article Search

JoP

Original Article

Split Viewer

Article

Original Article

J Pharmacopuncture 2024; 27(4): 358-366

Published online December 31, 2024 https://doi.org/10.3831/KPI.2024.27.4.358

Copyright © The Korean Pharmacopuncture Institute.

Assessing Hwa-byung Vulnerability Using the Hwa-byung Personality Scale: a comparative study of machine learning approaches

Chan-Young Kwon1* , Boram Lee2 , Sung-Hee Kim3 , Seok Chan Jeong4 , Jong-Woo Kim5

1Department of Oriental Neuropsychiatry, College of Korean Medicine, Dong-Eui University, Busan, Republic of Korea
2KM Science Research Division, Korea Institute of Oriental Medicine, Daejeon, Republic of Korea
3Department of Industrial ICT Engineering, Dong-Eui University, Busan, Republic of Korea
4Grand ICT Research Center, Department of e-Business and Graduate School of Artificial Intelligence, Dong-Eui University, Busan, Republic of Korea
5Department of Korean Neuropsychiatry, Kyung-Hee University Hospital at Gangdong, Seoul, Republic of Korea

Correspondence to:Chan-Young Kwon
Department of Oriental Neuropsychiatry, College of Korean Medicine, Dong-Eui University, 52-57 Yangjeong-ro, Busanjingu, Busan 47227, Republic of Korea
Tel: +82-51-850-8808
E-mail: beanalogue@deu.ac.kr

Received: October 18, 2024; Revised: November 11, 2024; Accepted: November 19, 2024

This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Objectives: To develop and compare machine learning models to classify individuals vulnerable to Hwa-byung (HB) using an existing HB personality scale and to evaluate the efficacy of these models in predicting HB vulnerability.
Methods: We analyzed data from 500 Korean adults (aged 19-44) using HB personality and symptom scales. We used various machine learning techniques, including the random forest classifier (RFC), XGBoost classifier, logistic regression, and their ensemble method (RFC-XGC-LR). The models were developed using recursive feature elimination with crossvalidation for feature selection and evaluated using multiple performance metrics, including accuracy, precision, recall, specificity, and area under the receiver operating characteristic curve (AUROC).
Results: The 16 items on the HB personality scale were identified as optimal features to predict high HB symptom scores requiring further clinical evaluation. The ensemble model slightly outperformed the other models, with an accuracy of 0.80 and an AUROC of 0.86, in the test set. Notably, item 16 (“I often feel guilty easily ”) of the HB personality scale showed the greatest importance in predicting HB vulnerability across all models. Although all models showed consistent performance across training, validation, and test sets, the RFC model exhibited signs of slight overfitting, with a higher AUROC of 0.97 in the training dataset compared to 0.85 in the validation and 0.86 in the test datasets.
Conclusion: Machine learning models, particularly the ensemble method, show capabilities promising for screening individuals with high HB symptom scores based on personality traits, potentially facilitating early referral for clinical evaluation. These models can improve the efficiency and accuracy of the HB risk assessment in clinical settings, potentially aiding early intervention and prevention strategies.

Keywords: Hwa-byung, Hwa-byung scale, machine learning, Korean medicine

INTRODUCTION

Hwa-byung (HB), also known as Korean “anger syndrome,” is a culture-bound syndrome primarily observed in Korean populations [1]. Traditionally, this mental health condition has been most common in middle-aged Korean women, with an epidemiological survey reporting a prevalence of 4.95% in this demographic [2]. HB is characterized by a combination of somatic and psychological symptoms and is often associated with the suppression of anger and emotional distress [1, 3]. The syndrome is closely associated with the Korean Confucian culture; however, at the individual level, it is associated with certain temperament and personality traits, such as low self-directedness and high anticipatory anxiety [4]. Recent studies have revealed the genetic vulnerability of an individual to HB [5]. Early identification of individuals vulnerable to HB is crucial for implementing timely interventions and preventing the progression of this syndrome.

The HB scale, developed in 2008 by Kwon et al. [5], consists of personality and symptom scales for assessing HB, and the personality scale evaluates the degree of personality vulnerability to HB using a scoring system. This scale serves as a valuable tool for evaluating an individual’s predisposition to HB [5]. However, unlike the symptom scale, the personality scale does not have a defined cut-off score [5], limiting its applicability in clinical settings. To date, research on its clinical value remains limited. However, a study focusing on Korean adolescents and young adults found that genetics significantly influenced HB personality [6], suggesting the biological validity of HB personality and highlighting its potential clinical value.

One of the challenges in studying HB personality lies in addressing the complex nature of HB and the multifaceted factors that contribute to its development. Machine learning techniques offer the potential to enhance the predictive power of existing assessment tools by identifying subtle patterns and their interactions [7, 8]. A significant advantage of machine learning-based analysis is its ability to analyze datasets involving multifaceted factors, such as personality traits, to develop optimized prediction models [8]. For example, Cho et al. [8] developed a machine learning-based predictive model that examines demographics, health data, and premorbid personality traits to predict the subsequent development of behavioral and psychological symptoms of dementia. However, no attempt has yet been made to develop a predictive model that would extend the clinical utility of traits of HB personality using this approach.

Recent studies have shown that the MZ generation in the Republic of Korea faces unique mental health challenges due to social pressure, economic uncertainty, and difficulties in achieving work-life balance [9-12]. Interestingly, a recent study revealed that 45.2% of the MZ generation exhibited reactive embitterment, a condition characterized by persistent feelings of inadequacy, a desire for revenge after experiencing insults or life stressors, and a sense of helplessness [10]. Given that these psychological characteristics closely align with the core features of HB, such as suppressed anger and emotional distress [3], the MZ generation may be particularly vulnerable to developing HB.

Therefore, this study used machine learning algorithms to develop a robust classification model to identify individuals in the MZ generation in the Republic of Korea who are at risk of developing HB. By comparing different modeling approaches, we attempted to determine the most effective method to predict HB vulnerability based on the results of the HB personality scale.

MATERIALS AND METHODS

1. Participants

This study analyzed a dataset of 500 adult respondents born between 1980 and 2005 (aged 19-44 years), considered the MZ generation. The participants were Korean citizens from the general population. Data were collected by Macromill Embrain (Embrain Co., Ltd., Seoul, Republic of Korea) through an anonymous online survey conducted between June 7 and June 12, 2024.

2. Sample size

The sample size was determined using demographic statistics and established sampling methodologies. The Korean Statistical Information Service reported that the MZ generation in the Republic of Korea comprised approximately 16.3 million individuals as of 2020 [13]. For populations larger than 1 million, Serdar et al. [14] proposed a sampling framework that recommended a minimum sample size of 384 participants to achieve statistical validity with a 5% margin of error. Based on this framework, we initially targeted a sample size of 384 participants and ultimately collected data from 500 participants to ensure robust statistical analysis.

3. Variables

Demographic data and clinical variables, including HB scores, were collected from the participants. The data included sex, which is a known risk factor for HB [1-3]. The HB scale comprises two components: the HB symptom scale and the HB personality scale. The symptom scale consists of 15 items, with participants self-reporting somatic and psychological symptoms of HB on a Likert scale ranging from 0 to 4 [5]. A total score of 30 or higher on the HB symptom scale [5] has been used as a screening threshold in previous clinical studies [15-17]. This threshold served as the target variable in our classification models, identifying individuals who may benefit from further clinical evaluation. The HB personality scale consists of 16 items, with participants self-reporting personality traits associated with vulnerability to HB on a Likert scale ranging from 0 to 4 [5]. As described earlier, a definitive cut-off value for the HB personality scale has not yet been developed.

4. Data analysis

Our analysis used a comprehensive machine learning approach to develop and evaluate classification models to predict HB vulnerability. Specifically, the HB screening threshold (i.e., an HB symptom scale score of ≥ 30) was used as the target variable. The primary aim was to develop a predictive model for HB based on the results of the HB personality scale. These steps were followed for the analysis (Fig. 1).

Figure 1. Process of this study. HB, Hwa-byung; ML, machine learning.
1) Data preprocessing

The dataset was thoroughly examined for missing values across all variables. No missing values were observed, ensuring the completeness and reliability of the data for subsequent analysis.

2) Feature selection

Recursive feature elimination with cross-validation (RFECV) was employed to identify the optimal subset of features for predicting HB. RFECV evaluates the performance of a model by gradually eliminating the less important features, thereby identifying key variables and optimizing the performance of the model [18]. For this analysis, feature selection was performed using a random forest classifier (RFC) as the base estimator [18]. A 5-fold cross-validation approach was applied, with the area under the receiver operating characteristic curve (AUROC) as the scoring metric. Through this process, important HB personality variables were selected to predict the HB screening threshold described above.

3) Data splitting

The dataset was divided into training (60%), validation (20%), and testing (20%) datasets. This approach helps prevent overfitting and enhances the reliability of performance evaluation results, a method commonly used in existing studies [19, 20]. Stratified sampling was used to ensure a balanced distribution of the target variable (i.e., the HB screening threshold) across all subsets.

4) Model development

We developed four classification models and compared their performance to develop an optimal prediction model [18, 19]: RFC, XGBoost classifier (XGC), logistic regression (LR), and voting classifier, an ensemble of the above three models (i.e., RFC-XGC-LR). The random forest is an ensemble method that constructs multiple decision trees, which are advantageous for identifying the importance of variables. XGBoost is a high-performance boosting algorithm designed to handle complex nonlinear relationships effectively. LR assumes a linear relationship between the features and the outcome, making it particularly strong for clinical applications. Finally, the voting classifier (i.e., RFC-XGC-LR) combines the predictions of RFC, XGC, and LR. By aggregating the strengths of individual models, the voting classifier aims to improve the overall prediction performance.

To optimize the hyperparameters of the RFC and XGC models, we used RandomizedSearchCV with 20 iterations and 5-fold cross-validation, using the AUROC as a scoring metric. Each model was trained using the training set and evaluated using a validation set. The final evaluation was performed on the test set to assess generalization. Separate evaluations for the training, validation, and test sets were performed to mitigate the risk of overfitting. The model performance was assessed using various metrics, including accuracy, precision, recall (sensitivity), specificity, AUROC, and average precision score. This score was obtained by calculating the precision and recall for various thresholds, drawing a precision–recall curve, and integrating the area under the curve.

After training the models, we generated a probability score between zero and one to predict the likelihood of HB vulnerability for each individual in the dataset, which we referred to as the HB vulnerability score. For each model, we determined the optimal classification threshold that maximized the difference between the true-positive and false-positive rates on the ROC curve. Finally, we performed a 10-fold cross-validation for each model to ensure robust performance estimates.

5. Software

The analysis was performed using Python 3.10.10 with the following libraries: pandas and numpy for data manipulation; scikit-learn for machine learning models, cross-validation, and evaluation metrics; XGBoost for implementing the XGBoost classifier; matplotlib for data visualization, and joblib for model serialization.

6. Ethical consideration

The study protocol was approved by the Institutional Review Board of the Dong-Eui University Korean Medicine Hospital (DH-2024-09).

RESULTS

1. General characteristics of participants

Of the 500 participants, 237 (47.4%) were female. The mean age of the participants was 34.74 ± 7.43 (median, 36) years. The mean HB personality score was 33.60 ± 9.32, with a median score of 33 points, while the mean HB symptom score was 26.67 ± 12.14, with a median score of 26 points. Notably, 195 participants (39%) scored above the established screening threshold (≥ 30) on the HB symptom scale, indicating a potential need for further clinical evaluation (Table 1).

HB, hwa-byung..

*A total score of 30 or more on the HB symptom scale..

&md=tbl&idx=1' data-target="#file-modal"">Table 1

Characteristics of included participants.

CharacteristicsMale (n = 263)Female (n = 237)Total (n = 500)
Mean age (years)35.43 ± 7.3933.98 ± 7.4334.74 ± 7.43
Median age (years)383536
Mean HB personality score1.93 ± 0.931.82 ± 0.801.88 ± 0.87
Mean HB symptom score26.79 ± 12.5526.54 ± 11.6826.67 ± 12.14
Above HB screening threshold* (n, %)101, 38.4%94, 39.7%195, 39.0%

HB, hwa-byung..

*A total score of 30 or more on the HB symptom scale..



2. Feature selection

For the RFC and XGC models, the importance of each feature was determined using the mean decrease in impurities. In the RFC-XGC-LR ensemble, feature importance was calculated as the average of the importance scores from the RFC and XGC models. For the LR model, feature importance was evaluated based on the absolute values of the regression coefficients. Specifically, item 16 of the HB personality scale consistently showed the highest importance across all models. Item 8 ranked second in importance in the RFC, XGC, and ensemble models, while item 12 ranked second in the LR model but showed lower importance in the other models (Fig. 2).

Figure 2. RFECV results for each model. RFECV, recursive feature elimination with cross-validation.

3. Model development

The four models, RFC (accuracy = 0.78, AUROC = 0.86), XGC (accuracy = 0.76, AUROC = 0.85), LR (accuracy = 0.78, AUROC = 0.85), and RFC-XGC-LR (accuracy = 0.80, AUROC = 0.86), showed strong predictive performance, and the ensemble model slightly outperformed the others on the test set (Fig. 3). The optimal classification thresholds varied between the models, ranging from 0.36 (LR) to 0.50 (RFC and RFC-XGC-LR). In RFC and RFC-XGC-LR, 39 out of 100 individuals (39%) in the test dataset scored above the cutoff. In XGC and LR, 49 participants scored above the cutoff (49%) (Fig. 4). The models achieved a good balance between sensitivity and specificity, with RFC-XGC-LR achieving the highest specificity (0.84) while maintaining a good sensitivity (0.74). All models showed consistent performance in training, validation, and test sets, indicating good generalization. However, there were signs of slight overfitting in the RFC, which showed higher performance in the training set (AUROC = 0.97) compared to the validation (AUROC = 0.85) and test (AUROC = 0.86) sets (Table 2). Finally, the 10-fold cross-validation revealed consistent performance across all models: RFC (mean AUROC = 0.82 ± 0.13), XGC (mean AUROC = 0.82 ± 0.13), LR (mean AUROC = 0.84 ± 0.15), and RFC-XGC-LR (mean AUROC = 0.83 ± 0.13).

AUROC, area under the receiver operating characteristic curve; Test, test dataset; Train, training dataset; Val, validation dataset..

&md=tbl&idx=2' data-target="#file-modal"">Table 2

Performance metrics for each model.

ModelOptimal thresholdsAccuracyPrecisionRecall (sensitivity)SpecificityAUROCAverage precision score
Random forestTrain: 0.41
Val: 0.36
Test: 0.50
Train: 0.91
Val: 0.79
Test: 0.78
Train: 0.84
Val: 0.67
Test: 0.72
Train: 0.96
Val: 0.92
Test: 0.72
Train: 0.88
Val: 0.70
Test: 0.82
Train: 0.97
Val: 0.85
Test: 0.86
Train: 0.96
Val: 0.76
Test: 0.82
XGBoostTrain: 0.40
Val: 0.34
Test: 0.40
Train: 0.85
Val: 0.78
Test: 0.76
Train: 0.77
Val: 0.65
Test: 0.65
Train: 0.88
Val: 0.92
Test: 0.82
Train: 0.83
Val: 0.69
Test: 0.72
Train: 0.93
Val: 0.85
Test: 0.85
Train: 0.89
Val: 0.77
Test: 0.82
Logistic
regression
Train: 0.34
Val: 0.29
Test: 0.36
Train: 0.79
Val: 0.79
Test: 0.78
Train: 0.69
Val: 0.67
Test: 0.67
Train: 0.83
Val: 0.92
Test: 0.85
Train: 0.76
Val: 0.70
Test: 0.74
Train: 0.87
Val: 0.86
Test: 0.85
Train: 0.81
Val: 0.79
Test: 0.80
Voting classifierTrain: 0.37
Val: 0.44
Test: 0.50
Train: 0.85
Val: 0.80
Test: 0.80
Train: 0.75
Val: 0.71
Test: 0.74
Train: 0.92
Val: 0.82
Test: 0.74
Train: 0.80
Val: 0.79
Test: 0.84
Train: 0.93
Val: 0.87
Test: 0.86
Train: 0.90
Val: 0.78
Test: 0.82

AUROC, area under the receiver operating characteristic curve; Test, test dataset; Train, training dataset; Val, validation dataset..


Figure 3. ROC curves indicating predictive performance for each model. AUC, area under the curve; ROC, receiver operating characteristic.
Figure 4. Distribution of HB vulnerability scores and threshold values for each model. HB, Hwa-byung.

DISCUSSION

Our findings showed the efficacy of machine learning approaches in predicting HB vulnerability based on the HB personality scale. The high performance of all the models, particularly the RFC-XGC-LR ensemble model, suggests that these approaches can reliably identify individuals at risk of developing HB. This has potential clinical significance, as early identification of vulnerable individuals could facilitate timely interventions and improve patient outcomes, reducing the burden of HB.

The relatively high proportion of participants scoring above the screening threshold (39%) in our study warrants careful interpretation. This rate may reflect the unique characteristics of our target population (i.e., the MZ generation), who have shown a high prevalence of related psychological conditions such as reactive embitterment (45.2%) [10]. Given that embitterment shares several psychological features with HB, including suppressed anger and emotional distress [3], this generation might be particularly vulnerable to HB-like symptoms. However, methodological factors such as online survey format, potential selection bias, and the use of screening thresholds rather than diagnostic criteria should also be considered when interpreting these findings.

The selection of all 16 items by the RFECV indicates that each trait contributes meaningful information to the prediction of HB vulnerability. This highlights the complexity of HB and the importance of considering a wide range of personality factors in its assessment [5]. In the RFECV results, item 16 (“I often feel guilty easily”) emerged as the most important predictor of HB vulnerability in all models. Guilt, along with shame, has been extensively studied for its association with mental health problems such as depression, anger, and aggression. Interestingly, guilt proneness is negatively related to outward anger and positively related to anger control, a tendency particularly pronounced in females [21]. This finding is consistent with the pathogenesis of HB, which attributes the condition to excessive suppression or internalization of anger [3]. Similarly, item 8 (“I often get disappointed in myself”) ranked second in importance for predicting HB in the three models: FC, XGC, and RFC-XGC-LR models. Self-disappointment can be seen as a form of self-criticism that threatens emotional well-being through rumination of anger [3]. Moreover, self-criticism can increase the risk of HB, as it involves feelings of guilt or anger [22]. However, the importance of item 8 was relatively low in the LR model. This is because LR assumes a linear relationship between features, which limits its ability to capture complex, nonlinear relationships that are better handled by models like RFC and XGC.

The slight superiority of the RFC-XGC-LR ensemble model in terms of accuracy and specificity highlights the potential benefits of ensemble methods in predicting HB vulnerability. Specifically, the high AUROC values across all models, particularly for RFC-XGC-LR (0.86), suggest that these models have strong discriminative power in identifying individuals at risk for HB. Furthermore, the consistent performance observed across cross-validation folds and different data splits (training, validation, and testing) indicates that these models are likely to generalize well to new datasets. However, the slight overfitting observed in the RFC model in the training set suggests that careful monitoring and further regularization may be beneficial to enhance its applicability in real-world scenarios. These findings have important implications for the early identification and prevention of HB. By accurately identifying high-risk individuals, healthcare providers can implement targeted interventions and support strategies, potentially improving outcomes. Moreover, the use of machine learning models can enhance the efficiency and accuracy of HB risk assessment in clinical settings, paving the way for more personalized and effective care.

This study has several limitations. First, the relatively small sample size may limit the generalizability of the findings. Although machine learning models have shown promising predictive capabilities, a larger and more diverse dataset is required to validate these results in different populations and settings. Future studies should include broader demographics to enhance the robustness and generalizability of the model. Second, the absence of external validation in this study is a noteworthy limitation. Although cross-validation was performed to evaluate the model’s performance, the lack of an independent test set limits the ability to confirm the model’s effectiveness in real-world settings. Incorporating external datasets for validation would provide a more comprehensive assessment of the predictive capabilities of the model. Third, high HB symptom scores (i.e., above the screening threshold), which were considered a target variable, depended on the total score on the HB symptom scale. Our findings should be interpreted within the context of screening-level assessment rather than diagnostic confirmation. Since clinician-led diagnostic tools, such as the HB Diagnostic Interview Schedule [3], are the standard in clinical settings, future studies should aim to validate these models using more definitive diagnostic criteria to reliably confirm the presence of HB.

CONCLUSION

This study demonstrated the effectiveness of machine learning approaches in screening for HB vulnerability using personality traits. The ensemble model (RFC-XGC-LR) achieved strong predictive performance (accuracy = 80%, AUROC = 0.86), highlighting the potential utility of machine learning in clinical screening processes. Notably, guilt proneness emerged as the most significant predictor across all models, suggesting its potential value as a key screening marker. While these results represent screening-level assessments rather than diagnostic outcomes, they provide valuable insights for clinical practice by identifying individuals who may benefit from comprehensive evaluation. The strong predictive performance of our models suggests that personality traits, when analyzed using machine learning approaches, can offer meaningful information for HB risk assessment. Future research should prioritize external validation and real-world implementation to confirm the practical utility of these findings. Advancements in screening methodologies like these could ultimately enhance the early identification and intervention strategies for individuals at risk of developing HB.

AUTHORS’ CONTRIBUTIONS

Conceptualization: Chan-Young Kwon; Methodology: Chan-Young Kwon, Sung-Hee Kim and Seok Chan Jeong; Software: Chan-Young Kwon; Formal analysis: Chan-Young Kwon; Data curation: Chan-Young Kwon; Writing—original draft preparation: Chan-Young Kwon; Writing—review and editing: Chan-Young Kwon, Boram Lee, Sung-Hee Kim, Seok Chan Jeong, and Jong-Woo Kim; Visualization: Chan-Young Kwon; Supervision: Chan-Young Kwon; Funding acquisition: Chan-Young Kwon. All authors have read and agreed to the published version of the manuscript.

ETHICAL APPROVAL

This study protocol was approved by the Institutional Review Board of Dong-Eui University Korean Medicine Hospital (DH-2024-09).

DATA AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon reasonable request.

CONFLICTS OF INTEREST

Chan-Young Kwon has been an editorial board member of Journal of Pharmacopuncture since 2022 but has no role in the decision to publish this article. No other potential conflicts of interest relevant to this aricle were reported.

FUNDING

This work was partly supported by Innovative Human Resource Development for Local Intel-lectualization program through the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MIST) (IITP-2024-RS-2020-II201791, 50%), and by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: RS-2023-KH139364, 50%).

Fig 1.

Figure 1.Process of this study. HB, Hwa-byung; ML, machine learning.
Journal of Pharmacopuncture 2024; 27: 358-366https://doi.org/10.3831/KPI.2024.27.4.358

Fig 2.

Figure 2.RFECV results for each model. RFECV, recursive feature elimination with cross-validation.
Journal of Pharmacopuncture 2024; 27: 358-366https://doi.org/10.3831/KPI.2024.27.4.358

Fig 3.

Figure 3.ROC curves indicating predictive performance for each model. AUC, area under the curve; ROC, receiver operating characteristic.
Journal of Pharmacopuncture 2024; 27: 358-366https://doi.org/10.3831/KPI.2024.27.4.358

Fig 4.

Figure 4.Distribution of HB vulnerability scores and threshold values for each model. HB, Hwa-byung.
Journal of Pharmacopuncture 2024; 27: 358-366https://doi.org/10.3831/KPI.2024.27.4.358

Table 1 . Characteristics of included participants.

CharacteristicsMale (n = 263)Female (n = 237)Total (n = 500)
Mean age (years)35.43 ± 7.3933.98 ± 7.4334.74 ± 7.43
Median age (years)383536
Mean HB personality score1.93 ± 0.931.82 ± 0.801.88 ± 0.87
Mean HB symptom score26.79 ± 12.5526.54 ± 11.6826.67 ± 12.14
Above HB screening threshold* (n, %)101, 38.4%94, 39.7%195, 39.0%

HB, hwa-byung..

*A total score of 30 or more on the HB symptom scale..


Table 2 . Performance metrics for each model.

ModelOptimal thresholdsAccuracyPrecisionRecall (sensitivity)SpecificityAUROCAverage precision score
Random forestTrain: 0.41
Val: 0.36
Test: 0.50
Train: 0.91
Val: 0.79
Test: 0.78
Train: 0.84
Val: 0.67
Test: 0.72
Train: 0.96
Val: 0.92
Test: 0.72
Train: 0.88
Val: 0.70
Test: 0.82
Train: 0.97
Val: 0.85
Test: 0.86
Train: 0.96
Val: 0.76
Test: 0.82
XGBoostTrain: 0.40
Val: 0.34
Test: 0.40
Train: 0.85
Val: 0.78
Test: 0.76
Train: 0.77
Val: 0.65
Test: 0.65
Train: 0.88
Val: 0.92
Test: 0.82
Train: 0.83
Val: 0.69
Test: 0.72
Train: 0.93
Val: 0.85
Test: 0.85
Train: 0.89
Val: 0.77
Test: 0.82
Logistic
regression
Train: 0.34
Val: 0.29
Test: 0.36
Train: 0.79
Val: 0.79
Test: 0.78
Train: 0.69
Val: 0.67
Test: 0.67
Train: 0.83
Val: 0.92
Test: 0.85
Train: 0.76
Val: 0.70
Test: 0.74
Train: 0.87
Val: 0.86
Test: 0.85
Train: 0.81
Val: 0.79
Test: 0.80
Voting classifierTrain: 0.37
Val: 0.44
Test: 0.50
Train: 0.85
Val: 0.80
Test: 0.80
Train: 0.75
Val: 0.71
Test: 0.74
Train: 0.92
Val: 0.82
Test: 0.74
Train: 0.80
Val: 0.79
Test: 0.84
Train: 0.93
Val: 0.87
Test: 0.86
Train: 0.90
Val: 0.78
Test: 0.82

AUROC, area under the receiver operating characteristic curve; Test, test dataset; Train, training dataset; Val, validation dataset..


References

  1. Lin KM. Hwa-Byung: a Korean culture-bound syndrome? Am J Psychiatry. 1983;140(1):105-7.
    Pubmed CrossRef
  2. Park YJ, Kim HS, Kang HC, Kim JW. A survey of Hwa-Byung in middle-age Korean women. J Transcult Nurs. 2001;12(2):115-22.
    Pubmed CrossRef
  3. The Society of Korean Medicine Neuropsychiatry. Hwabyung: clinical practice guideline of Korean medicine. Gyeongsan: National Institute for Korean Medicine Development; 2021. p. 195.
  4. Lee J, Min SK, Kim KH, Kim B, Cho SJ, Lee SH, et al. Differences in temperament and character dimensions of personality between patients with Hwa-byung, an anger syndrome, and patients with major depressive disorder. J Affect Disord. 2012;138(1-2):110-6.
    Pubmed CrossRef
  5. Kwon JH, Park DG, Kim JW, Lee MS, Min SG, Kwon H. Development and validation of the Hwa-Byung Scale. Korean J Clin Psychol. 2008;27(1):237-52.
    CrossRef
  6. Hur YM. Genetic and environmental influences on Hwabyung-personality in South Korean adolescents and young adults. Stress. 2020;28(1):25-32.
    CrossRef
  7. Sun H, Depraetere K, Meesseman L, Cabanillas Silva P, Szymanowsky R, Fliegenschmidt J, et al. Machine learning-based prediction models for different clinical risks in different hospitals: evaluation of live performance. J Med Internet Res. 2022;24(6):e34295.
    Pubmed KoreaMed CrossRef
  8. Cho E, Kim S, Heo SJ, Shin J, Hwang S, Kwon E, et al. Machine learning-based predictive models for the occurrence of behavioral and psychological symptoms of dementia: model development and validation. Sci Rep. 2023;13(1):8073.
    Pubmed KoreaMed CrossRef
  9. Kim J. Fairness in Korean society: assessing the perspective of millennials. Technium Soc Sci J. 2020;11(1):482-95.
    CrossRef
  10. Lee JH, Kim S. Exposure to negative life events and post-traumatic embitterment symptoms in young adults in Korea: cumulative and differential effects. Psychopathology. 2019;52(1):18-25.
    Pubmed CrossRef
  11. Moon S, Kim Y. Subjective perceptions of 'meaning of work' of generation MZ employees of South Korean NGOs. Behav Sci (Basel). 2023;13(6):461.
    Pubmed KoreaMed CrossRef
  12. Yim DH, Kwon Y. Does young adults' neighborhood environment affect their depressive mood? Insights from the 2019 Korean Community Health Survey. Int J Environ Res Public Health. 2021;18(3):1269.
    Pubmed KoreaMed CrossRef
  13. Lee D. Analyzing the Korean MZ generation [Internet]. Sejong: Ministry of Culture, Sports and Tourism; 2024 [cited 2024 Oct 10]. Available from: https://www.kocis.go.kr/koreanet/view.do?seq=1047885.
  14. Serdar CC, Cihan M, Yücel D, Serdar MA. Sample size, power and effect size revisited: simplified and practical approaches in pre-clinical, clinical and laboratory studies. Biochem Med (Zagreb). 2021;31(1):010502.
    Pubmed KoreaMed CrossRef
  15. Suh HW, Ko Y, Moon S, Kim JW, Chung SY, Hong S, et al. A multicenter registry of neuropsychiatric outpatients in Korean medicine hospitals (KMental): protocol of a prospective, multicenter, registry study. Medicine (Baltimore). 2022;101(49):e32151.
    Pubmed KoreaMed CrossRef
  16. Kim NS, Lee KE. Gender differences in factors affecting Hwa-byung symptoms with middle-age people. J Korean Acad Fundam Nurs. 2012;19(1):98-108.
    CrossRef
  17. Choi YH, Jin MK, Kim BK. A comparative study of communication type and stress coping style between Hwabyung patients group and non-Hwabyung patients group. J Orient Neuropsychiatry. 2015;26(4):365-81.
    CrossRef
  18. Liu W, Zhang L, Bao L, Shen G, Feng J. Accurate classification and prediction of acute myocardial infarction through an ARMD procedure. J Proteome Res. 2023;22(3):758-67.
    Pubmed CrossRef
  19. Chen KA, Berginski ME, Desai CS, Guillem JG, Stem J, Gomez SM, et al. Differential performance of machine learning models in prediction of procedure-specific outcomes. J Gastrointest Surg. 2022;26(8):1732-42.
    Pubmed KoreaMed CrossRef
  20. Madakkatel I, Zhou A, McDonnell MD, Hyppönen E. Combining machine learning and conventional statistical approaches for risk factor discovery in a large cohort study. Sci Rep. 2021;11(1):22997.
    Pubmed KoreaMed CrossRef
  21. Lutwak N, Panish JB, Ferrari JR, Razzino BE. Shame and guilt and their relationship to positive expectations and anger expressiveness. Adolescence. 2001;36(144):641-53.
  22. Austin J, Drossaert CHC, Sanderman R, Schroevers MJ, Bohlmeijer ET. Experiences of self-criticism and self-compassion in people diagnosed with cancer: a multimethod qualitative study. Front Psychol. 2021;12:737725.
    Pubmed KoreaMed CrossRef