A MLP based predictive model for risk assessment of workplace violence for emergency nurses in China

Workplace violence (WPV) is a critical occupational hazard, particularly for emergency department nurses who work in high-stress and unpredictable environments. Emerging evidence suggests that intrinsic personality traits, such as psychological resilience and moral courage, are associated with WPV vulnerability.

We conducted a nationwide study involving 465 emergency nurses from 13 provincial hospitals in China. The data were collected using validated instruments, including CD-RISC, NMCS, DASS-21, NAQ-R, and WVS. Unsupervised clustering was applied to identify personality trait subgroups with distinct WPV profiles. A Multi-Layer Perceptron model was then developed using only demographic features to predict WPV risk, with logistic regression, random forest, support vector machine, and XGBoost models used as comparators.

Three personality clusters were identified, with one subgroup exhibiting significantly higher WPV scores (mean = 3.51). This high-risk group, characterized by lower resilience and moral courage, was conceptualized as a "nursing Persona" for targeted intervention. The MLP model trained on demographic data achieved the highest performance in cross-validation (mean accuracy = 0.711) compared to other models. External validation confirmed the model's generalizability, distinguishing high-risk nurses with significantly elevated WPV scores (mean = 4.344 vs. 2.504, p = 0.001).

We identified a high-risk nurse Persona vulnerable to WPV through unsupervised clustering of personality trait. A robust and generalizable MLP model was developed to stratify WPV risk using accessible demographic features, offering a practical tool for early identification and targeted intervention in occupational health settings.