Social-psychological Dual-dimensional Clustering Modeling: Heterogeneity Analysis of Smoking Cessation Success Based on Machine Learning

Smoking poses a serious threat to public health worldwide, with significant differences in smoking behavior and quit success rates among different occupational groups. We propose an "Occupational Clustering - Dynamic Modeling" framework, which involves finely classifying occupations into seven major categories and analyzing them from both the "social-psychological" dimensions to construct a differentiated intervention model. From a methodological standpoint, we apply Principal Component Analysis (PCA) to perform dimensionality reduction and extract the most informative features. K-Means clustering is employed to reveal latent socio-psychological groupings within each occupational category, thereby enabling a stratified examination of populations. Random Forest classifier is built and assessed across multiple combinations of input variables: the configuration yielding the highest validation performance is chosen, and its feature‐importance metrics pinpoint the principal drivers of quitting success. To rectify the imbalance between cessation outcomes, the Borderline-SMOTE oversampling approach is used, which enhances the model's ability to detect and correctly classify minority‐class instances. We find significant differences in quit success rates and feature influences among different occupational groups, with the group model performing better in specific occupational groups and the Random Forest model accurately identifying key influencing factors. The use of grouping-based smoking cessation prediction models and feature importance analysis helps in formulating precise intervention strategies, improving quit success rates, advancing the field of smoking cessation interventions, and enhancing public health.