MTA-CPMA: multi-tier architecture with cross-platform mobile apps for student health and academic analytics

Background Mobile health (mHealth) applications have predominantly focused on isolated health behaviours, neglecting holistic wellness approaches and their potential correlation with academic performance. Existing educational portals report only academic performance and attendance records, failing to capture health status indicators such as dietary habits, physical activity, sleep quality, and psychological well-being. A unified platform integrating health behaviour monitoring and academic performance analytics is therefore needed in university environments. Methods This study proposes a Multi-Tier Architecture with Cross-Platform Mobile Apps (MTA-CPMA) for a Mobile App-Based Monitoring and Intelligent Intervention System (MAMIS), designed for university settings. MTA-CPMA integrates health behaviour monitoring with academic performance analytics within a requirements-driven layered framework serving students, faculty/staff, and healthcare providers. MAMIS employs a multi-model ensemble combining Long Short-Term Memory (LSTM) networks for temporal pattern analysis, Random Forest classifiers for risk assessment, Gradient Boosting for health-academic correlation analysis, and Isolation Forest for anomaly detection, with risk prediction via confidence-weighted ensemble fusion. A four-level intervention protocol (Preventive, Behavioural Nudge, Active Support, Escalated Response) delivers graduated, personalised health recommendations with machine learning-driven timing optimisation. The architecture was evaluated using a CloudSim Plus simulation environment with augmented training data ( n ≈ 2,500 student records) from original health and academic data. Results Rather than applying classical significance testing to synthetic data, a simulation-appropriate robustness framework was employed. Multi-seed stability analysis (30 seeds) yielded mean accuracy 89.3 ± 1.4% and area under the receiver operating characteristic curve (AUC-ROC) 0.923 ± 0.012. Noise sensitivity testing retained 85.1% accuracy at 10% injected noise; missingness robustness under missing completely at random (MCAR) conditions retained 84.7% at 20% missing data. Correlation recovery analysis achieved a mean recovery rate of 93.0% for known ground-truth health-academic correlations. The ensemble outperformed the best individual classifier by 5.4 percentage points. Conclusions This simulation-validated framework demonstrates the architectural feasibility and machine-learning predictive capability of an integrated health-academic monitoring platform. MTA-CPMA provides a scalable, robust foundation for deploying personalised health interventions in university environments, with results supporting its potential to proactively address student well-being alongside academic outcomes. Real-world validation with live student populations is required before findings can be generalised beyond the simulation context.