Comparative Analysis of Machine Learning Algorithms for Predicting Difficulty Disengaging from Social Media Among Students
Abstract
Social media use can adversely affect students’ academic functioning and psychological well-being when disengagement becomes difficult. This study compared Random Forest (RF), XGBoost, and Support Vector Machine (SVM) for predicting self-reported difficulty disengaging from social media among university students. A cross-sectional survey dataset comprising 118 responses was analyzed, including 75 positive (63.6%) and 43 negative (36.4%) cases. The binary target represented a behavioral proxy rather than a clinically validated diagnosis of social media addiction. Using balanced class weighting and 10-fold stratified cross-validation, SVM achieved the highest mean accuracy (77.2%) and ROC-AUC (0.804), followed by RF (73.0%) and XGBoost (66.8%). However, the performance difference between SVM and RF was not statistically significant (Nadeau–Bengio corrected *p* = 0.102), while RF demonstrated greater resampling stability. The findings provide preliminary evidence for algorithm selection in behavioral proxy classification and support future digital well-being research. Interpretation is limited by the small, single-institution sample and single-item outcome measure; therefore, the results should not be used for clinical screening or institutional deployment without external validation using established psychometric instruments.
Related
- Corrections to ``The Method of Gaps: Exact Expressions for the Generalization Error of Supervised Learning Algorithms
- Optimization Frameworks & Learning Algorithms for Discrete-Time Mean Field Games
- Mean Field Reinforcement Learning: Theory and Algorithms for Mean Field Control
- From Privacy to Generalization: Linear Max-Information Bounds for Differentially Private Learning Algorithms
- Parameter estimation in Conditional Sequential Monte Carlo algorithms through Particle Learning
BibTeX
@article{supardi2026comparative,
title = {{Comparative Analysis of Machine Learning Algorithms for Predicting Difficulty Disengaging from Social Media Among Students}},
author = {Supardi and Djoko Soetarno and Rahmat Sulaiman and Chandra Kirana and Ari Amir Alkodri and Anisah},
journal = {Journal of Information Systems and Informatics},
year = {2026},
month = oct,
doi = {10.63158/journalisi.v8i5.1735},
url = {https://doi.org/10.63158/journalisi.v8i5.1735},
note = {CC BY},
}
RIS
TY - JOUR
TI - Comparative Analysis of Machine Learning Algorithms for Predicting Difficulty Disengaging from Social Media Among Students
AU - Supardi
AU - Djoko Soetarno
AU - Rahmat Sulaiman
AU - Chandra Kirana
AU - Ari Amir Alkodri
AU - Anisah
PY - 2026
DA - 2026/10/24
JO - Journal of Information Systems and Informatics
DO - 10.63158/journalisi.v8i5.1735
UR - https://doi.org/10.63158/journalisi.v8i5.1735
AB - Social media use can adversely affect students’ academic functioning and psychological well-being when disengagement becomes difficult. This study compared Random Forest (RF), XGBoost, and Support Vector Machine (SVM) for predicting self-reported difficulty disengaging from social media among university students. A cross-sectional survey dataset comprising 118 responses was analyzed, including 75 positive (63.6%) and 43 negative (36.4%) cases. The binary target represented a behavioral proxy rather than a clinically validated diagnosis of social media addiction. Using balanced class weighting and 10-fold stratified cross-validation, SVM achieved the highest mean accuracy (77.2%) and ROC-AUC (0.804), followed by RF (73.0%) and XGBoost (66.8%). However, the performance difference between SVM and RF was not statistically significant (Nadeau–Bengio corrected *p* = 0.102), while RF demonstrated greater resampling stability. The findings provide preliminary evidence for algorithm selection in behavioral proxy classification and support future digital well-being research. Interpretation is limited by the small, single-institution sample and single-item outcome measure; therefore, the results should not be used for clinical screening or institutional deployment without external validation using established psychometric instruments.
SN - 2656-5935
SN - 2656-4882
ER -