Predicting the Impact of Mobile Phone Usage on Academic Performance Using Machine Learning: A Real-Time Digital Wellness Approach

This study investigates how mobile phone usage affects students’ academic performance using advanced machine learning methods, with the goal of enabling real-time digital wellness risk detection and targeted educational interventions. Data from 5,000 students (2023–2024) were collected via surveys, academic records, and digital behavior tracking, capturing variables such as screen time, app categories, study-to-phone usage ratio, and sleep duration. The authors implemented a CNN–LSTM architecture with an attention mechanism alongside traditional models including Random Forest, Gradient Boosting, Naive Bayes, SVM, and KNN. The CNN–LSTM with attention achieved 92% predictive accuracy, outperforming all baseline models. Findings show that using non-educational apps for more than 4 hours per day is associated with a 20% decline in academic performance, and the study-to-phone usage ratio is the most influential predictor. Educational app usage exhibited only a marginal positive relationship with performance. The results demonstrate that machine learning can effectively identify at-risk students and support the design of real-time monitoring frameworks and evidence-based mobile phone usage policies in educational settings.