Wild mushroom consumption susceptibility among Chinese university students: A machine learning study

Objectives: To investigate factors associated with susceptibility to wild mushroom consumption using machine learning approaches and identify key predictors for targeted intervention development. Methods: A cross-sectional survey of 216 Chinese university students employed three machine learning algorithms (Logistic Regression, Random Forest, Extremely Randomized Trees [ExtraTrees]) to predict consumption susceptibility based on demographics, media usage, and cognitive factors. Susceptibility was assessed through scenario-based questions following established frameworks from tobacco research. Model performance was evaluated using AUC with 95% confidence intervals calculated via bootstrap resampling (1,000 iterations). Sensitivity analyses were conducted using alternative susceptibility thresholds. Results: 65.3% were classified as susceptible to consumption. Logistic Regression achieved highest performance (AUC = 0.776, 95% CI: 0.679–0.862). Risk perception emerged as the strongest predictor (importance = 0.133 ± 0.044), followed by mushroom picking experience (0.101 ± 0.017) and content impression (0.089 ± 0.018). Among the 63 participants (29.2%) who reported using AI models, 75.93% indicated trust levels of ‘fairly trust’ or above. Conclusions: In this exploratory study of Chinese university students from a single institution, cognitive factors, particularly risk perception and identification ability, showed the strongest associations with consumption susceptibility. These preliminary findings suggest that targeted interventions enhancing risk awareness may be relevant for this population, though replication across diverse samples is needed before broader conclusions can be drawn.