Detecting internet addiction risk in vocational college students: Based on machine learning approach and Latent Profile Analysis

Internet Addiction (IA) is a significant public health concern, particularly among vocational college students who navigate unique stressors. This study adopted an innovative approach by integrating machine learning (ML) with Latent Profile Analysis (LPA) to identify key predictors of IA and uncover heterogeneity within at-risk populations. Combined with SHAP analysis to enhance the interpretability of machine learning. Based on a large-scale sample of 1,177 vocational college students, three ML models were trained on groups of the top and bottom 27% of IA scorers. Across all models, factors related to psychological distress, including emotional exhaustion, anxiety, depression, and stress, consistently emerged as the most critical predictors of IA risk. An LPA conducted on the middle-risk subsample (middle 46% of IA scorers) revealed three distinct subgroups: "High Distress and High Self-Esteem", "Moderate Distress and High Self-Esteem" , and "Low distress and high well-being". Subsequent risk predictions from the trained ML models externally validated this classification, confirming that the high-distress and moderate-distress profiles were significantly more vulnerable to IA than the healthy profile. These findings indicate that the "middle-risk" population is not uniform and highlight two noteworthy subgroups characterized by a combination of high distress and high self-esteem (possibly defensive). By combining ML and LPA, this study provides a powerful, nuanced framework for identifying specific at-risk students, thereby it provides the idea for targeted and personalized interventions.