A review of digital addiction as a predictor in ML-based academic performance prediction

Abstract

The rapid growth of digital technologies has raised growing concerns about digital addiction and its potential impact on academic performance of students. Although machine learning (ML) techniques are widely employed to predict academic outcomes, the systematic integration of digital addiction-related variables within predictive modeling frameworks remains conceptually and methodologically inconsistent. This systematic review aims to synthesize existing research on digital addiction as a predictive factor in ML-based academic performance prediction by examining digital addiction measures, predictive features, datasets, machine learning approaches, and methodological trends. Following the PRISMA 2020 guidelines, a systematic search was conducted across IEEE Xplore, ACM Digital Library, ScienceDirect, and Google Scholar to identify peer-reviewed Q1, Q2, and Q3 journal articles published between 2017 and 2025. After applying predefined inclusion and exclusion criteria, 41 studies were selected for qualitative synthesis. The review identified that Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Artificial Neural Networks (ANN), and ensemble learning methods were the most frequently adopted predictive models. Digital addiction was primarily measured using validated psychometric instruments and was increasingly incorporated alongside behavioral, demographic, and academic features to enhance predictive performance. However, substantial methodological heterogeneity, inconsistent operationalization of digital addiction, and the limited availability of standardized publicly accessible benchmark datasets restrict reproducibility, external validation, and cross-study comparability. Overall, the findings demonstrate the potential value of digital addiction-related variables to improve ML-based academic performance prediction while highlighting the need for standardized measurement frameworks, benchmark datasets, explainable machine learning approaches, and ethically responsible predictive modeling to support future educational research and practice.