Objective: To develop and compare multiple machine learning models for identifying high-risk college students exhibiting phubbing behavior and to determine key predictive factors, so as to provide evidence for precise screening and early intervention strategies. Methods: In December 2025, 1 828 undergraduate students, selected from three universities in Wuhan by using convenience sampling method, were surveyed via online qustionnaire for sociodemographic characteristics, phubbing behavior, family cohesion index, mature happiness, fear of negative evaluation, multidimensional state boredom, and psychological vulnerability. Four machine learning algorithms, including multilayer perceptron (MLP), extreme gradient booting (XGBoost), K-nearest neighbors (KNN), and gradient boosting decision tree (GBDT), were applied. The dataset was randomly split into a training set and a test set at a ratio of 7∶3. Model performance was evaluated using accuracy, recall, F1-score, and area under the receiver operating characteristic curve (AUC). The best performing model was further interpreted using shapley additive explanation (SHAP) analysis to assess feature importance. Results: In the training set,955 participants were classified as having low level phubbing behavior, and 325 participants were classified as having high level phubbing behavior. Significant differences were observed in psychological vulnerability, multidimensional state boredom, family cohesion index, and mature happiness between groups ( Z =-10.29, -11.72, -8.17, -7.83, all P <0.05). Significant differences were also found in the detection rate of high level phubbing behavior according to family residence, family status, left behind experience,interpersonal relationships,exercise frequency,physical flexibility,and sleep status ( χ 2=12.22, 38.93, 16.90, 44.64, 58.17, 82.14, 37.89, all P <0.05). Among the four machine learning models (MLP, KNN, GBDT, XGBoost),all showed good discrimination ability, with XGBoost performing best (accuracy 79%, recall 71%, F1-score=0.70, AUC=0.78), while GBDT achieved the highest precision (78%). Feature importance analysis showed that multidimensional state boredom was the most important predictor of phubbing behavior, followed by psychological vulnerability, family cohesion, mature happiness, and fear of negative evaluation. SHAP analysis indicated that multidimensional state boredom, psychological vulnerability, and fear of negative evaluation had positive risk effects, while family cohesion and hedonic well being had protective effects for high risk of phubbing behavior. Conclusions Multidimensional state boredom is the core predictor of phubbing behavior among college students. The XGBoost model demonstrates good predictive performance and can provide a reference for identifying high risk individuals and implementing targeted interventions.