Positive relationship between education level and risk perception and behavioral response: A machine learning approach

PLOS ONE | https://doi.org/10.1371/journal.pone.0321153 April 3, 2025 4 / 10 PLOS ONE Positive relationship between education level and risk perception and behavioral response at-risk students [11]. These applications highlight ML’s potential to reveal complex relation- ships and improve decision-making in risk-related contexts. Random Forest and XGBoost are ML algorithms used for regression and classification, both based on ensemble learning to improve performance by combining multiple models. Random Forest, an ensemble method built on decision trees, constructs numerous trees trained on random subsets of data [12]. Each tree makes independent predictions, which are aggregated through voting or averaging to yield the final prediction. This approach enhances robustness and accuracy while reducing overfitting. XGBoost (Extreme Gradient Boosting), based on gradient-boosted trees [13], iteratively builds decision trees that correct the residual errors of the previous tree, optimizing model performance by minimizing a loss function. Known for its high accuracy, efficiency, and ability to handle large datasets, XGBoost excels in predictive tasks. This allows us to rank feature importance and uncover key influencing factors. By visual- izing feature SHAP values and their relationships with feature values, along with interaction plots, we can construct models that highlight the underlying trends. As shown in Table 1, behavioral intentions and preparatory behaviors serve as depen- dent variables, while 27 influencing factors are treated as independent variables to build

ML predictive models [4,7,14–22], including Random Forest Regression and XGBoost Table 1. Definition of Machine Learning Variables. Variable Type Variable Name Variable Symbol Variable Description Dependent Variable 1 Behavioral Intention mtotscore Factor Scores for Behavioral Intention Items Dependent Variable 2 Preparatory Behavior ytotscore Factor Scores for Preparatory Behavior Items Influencing Factors Risk Perception xtotscore Factor Scores for Risk Perception Items Age [14] Q11 What is your age? A. 18 or younger B. 19-35 C. 36-45 D. 46 ~ 60 E. 61 or older Education Level [15] Q12 What is your highest level of education? A. Junior high school or below B. High school or vocational school C. College or undergraduate D. Master’s degree or higher Household Type [16] Q13 What is your household type? A. Urban B. Rural Marital Status [15] Q14 What is your marital status? A. Single B. Married C. Divorced D. Widowed Health Status [4] Q16 How do you assess your health status? A. Very poor B. Poor C. Average D. Good E. Very good Occupation [17] Q17 What is your employment status? A. Pubic sector (e.g., teacher, doctor, police officer, judge, civil servant) B. Private sector (e.g., self-employed, private business owner, manual laborer) C. Unemployed Earthquake Experience [18] Q19 Have you ever experienced an earthquake? A. Yes B. No Housing Type [19] Q22 What is your housing type? A. Rented B. Owned C. Self-built D. Other (e.g., irregu- lar land usage) Satisfaction with the Timeliness of Local Treatment [7]

Q23 How satisfied are you with the timeliness of treatment near your residence? A. Very dissatisfied B. Dissatisfied C. Neutral D. Satisfied E. Very satisfied Province of Residence [20] Q24 Where is your residence located? [Open-ended] Information Access via Authorita- tive Media [15] Q25-1 How do you primarily obtain information about earthquake-related issues in your area? [Select all that apply] A. Television, news, and other authoritative media B. Social media (e.g., Weibo, WeChat) C. Propaganda from work or community organizations D. Other Information Access via Social Media [15] Q25-2 Information Access via Community Organizations [15] Q25-3 (Continued)