Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Machine learning for performance

Predicting college students’ exercise dependence: a machine learning approach

Deng Y; Lan W; Si M; Ren YL

Originally published in Front Psychol · 2026-01-29

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

AbstractExercise dependence behavior among college students is a critical issue in sports psychology that deserve closer examination, and artificial intelligence offer a useful ways to explore its mechanisms and predicting associated risks. In this study, data were collected from 2,745 college students using three standardized questionnaires, covering (i) exercise dependence behavior, (ii) psychological characteristics (e.g., exercise identity, weight biases), and (iii) basic demographic information. We used four widely used machine learning algorithms: logistic regression, random forest, extreme gradient boosting (XGBoost), and multilayer perceptron, and their outputs were further integrated through an ensemble learning techniques to further enhance the robustness and predictive power of the models. The stacking ensemble model…

Classification accuracy and confusion matrix analysis results

Figure 6 shows the confusion matrices for the four models. Cells on the main diagonal represent correct classification; off-diagonal cells indicate errors. All four models distinguished the three risk groups effectively. Logistic regression and the multilayer perceptron (MLP) reached the highest accuracy (accuracy = 0.98), random forest reached 0.97, and XGBoost reached 0.96. Across the three risk groups, all models classified most students correctly. MLP and logistic regression had the highest correct counts, random forest was close, and XGBoost was slightly behind.

Conclusion

To the best of our knowledge, for the first time various artificial intelligence models are used to investigate exercise dependence in Chinese college students. Using data on 2,745 students, we examined prevalence trends and demographic correlates as well as the main predictors. 33% of the students exhibited risk signals, whereas males and lower class levels and normal BMI students had a higher proportion of risk. The AUC of the 4-classifier-based stacked model was very high (99.16%), allowing accurate risk stratification. The model-interpretation results demonstrated that for prediction, behavioral/psychological items were more informative than demographics. There are 4 of them: extending exercise time to achieve the desired effect (A17), spending maximum time with leisure activity-exercising way (A13), not able to reduce frequency…

Editorial reading guide

  • What data created and tested the model, and was evaluation independent?
  • What is the cost of a false positive, false negative or out-of-distribution decision?