Journal of Engineering Research and Reports

The study aimed to quantify the influence of students’ mobile phone behaviors—including daily screen time, non‐educational versus educational app usage, study‐to‐phone ratios, and nighttime device checks—on academic performance and to develop a real‐time digital wellness alert system for early identification of at‐risk learners. To achieve this, researchers collected digital activity logs and academic records from 5,000 consenting students across middle school, high school, and university during the 2023–24 academic year, engineered features such as screen time totals, app category hours, study‐to‐phone ratios, and sleep metrics, and then trained a hybrid deep learning model that combines convolutional neural networks, long short‐term memory units, and an attention mechanism. This CNN‐LSTM with attention was benchmarked against four traditional classifiers (Random Forest, Gradient Boosting, Support Vector Machine, and Naive Bayes), each optimized via grid search and validated through five‐fold cross‐validation, with all models evaluated on accuracy, precision, recall, F1‐score, and ROC‐AUC. The deep learning approach outperformed all baselines, achieving 92% accuracy and over 91% on every other metric compared to the best traditional model’s 88% accuracy—and revealed clear behavioral thresholds: over four hours of non‐educational app use corresponded to a 20% performance drop, while maintaining a study‐to‐phone ratio above 2 : 1 yielded a 15% grade improvement; heavy social media and gaming use led to declines of 14% and 16%, respectively, whereas educational app engagement produced a modest 3.5% boost. The novelty of this work lies in its integration of convolutional, sequential, and attention layers to detect critical usage spikes, the establishment of precise intervention benchmarks, and the demonstration of how embedding such predictive models into educational dashboards can shift student support from reactive remediation to proactive wellness promotion.