Feature Importance Ranking in Machine Learning Models Predicting Help-Seeking Intentions for Mental Health Issues in Women with High Public Self-Awareness

Abstract

Objective: The objective of this study was to utilize machine learning algorithms to predict professional psychological help-seeking intentions and robustly rank the importance of various psychosocial and demographic features among Taiwanese women exhibiting high public self-awareness. Methods and Materials: This cross-sectional study recruited N=1,452Taiwanese women identified with high public self-awareness via purposive sampling. Data were collected using validated instruments translated into Traditional Chinese, including the Self-Consciousness Scale, the Mental Help Seeking Intention Scale, and the Kessler Distress Scale. The analytical pipeline was executed using Python (Scikit-learn and XGBoost libraries) for data preprocessing, model training, and performance evaluation across four algorithms (Logistic Regression, Support Vector Machine, Random Forest, and XGBoost). Finally, SHapley Additive exPlanations (SHAP) were employed to extract and rank the importance of each feature in the optimal predictive model. Findings: The participant cohort presented a mean age of 31.8years, with 77.9%holding at least a bachelor’s degree, and 44.6%demonstrating high intentions to seek professional psychological help. The XGBoost algorithm achieved superior predictive performance on the test set compared to traditional models, yielding an Area Under the Curve (AUC) of 0.91. SHAP value analysis identified “Self-Stigma of Seeking Help” (Mean ∣SHAP∣=1.24) and “Perceived Public Stigma” (Mean ∣SHAP∣=0.98) as the most paramount negative predictors paralyzing help-seeking intentions, whereas “Mental Health Literacy” (Mean ∣SHAP∣=0.85) emerged as the strongest positive predictor mitigating these barriers. Conclusion: Internalized and public stigma profoundly override clinical need and demographic factors in highly self-aware women, necessitating the implementation of highly confidential, low-barrier mental health interventions that circumvent the fear of social evaluation.