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 6 / 10 PLOS ONE Positive relationship between education level and risk perception and behavioral response 3. Results 3.1 Model training and evaluation The dataset is split into training and validation sets in an 8:2 ratio. Model performance is evaluated using of MAE, MSE, and R2, as shown in Table 2. MAE and MSE measure the error between observed and predicted values, with smaller values indicating lower error. R2 reflects the goodness of fit, with higher values signifying better model performance. Based on these metrics, the random forest model demonstrates superior performance and will be used in sub- sequent experiments. Although the selected variables explain only 25.71% of the variance in the preparatory behav- ior model (ytotscore) and less than 20% in the behavioral intention model (mtotscore), falling short of fully accounting for the output variables in the random forest model, the primary aim of this section is to identify specific scenarios by examining internal influencing factors within a general context, rather than predicting the output variables. Thus, the goodness of fit serves only as a criterion for model selection and does not impede the exploration of specific scenarios in subsequent analyses. 3.2 Results description As shown in Fig 1, with preparatory behavior as the output variable, Fig 1(a) presents the global feature importance ranking based on the average absolute Shapley values. It’s clear that Table 2. Evaluation of Machine Learning Performance. MAE MSE R2

ytotscore RF 0.4861 0.3506 0.2571 XGB 0.5090 0.3951 0.1629 mtotscore RF 0.5287 0.4577 0.1419 XGB 0.5597 0.5277 0.0106 https://doi.org/10.1371/journal.pone.0321153.t002 Fig 1. Feature Impact Summary Based on SHAP Values. left: (a) Feature Importance Ranking; Right: (b) Influence of Features on Preparatory Behavior Effectiveness. https://doi.org/10.1371/journal.pone.0321153.g001