Neuro-Integrated Residual-Attention Learning for Robust Outcome Prediction in Mobile Remote Biology Laboratories

Abstract

Despite heterogeneous learner behaviors, interaction noise, and unstable session dynamics, AI-enabled remote biology laboratories are increasingly used to support scalable and flexible science education. Assessing learning outcomes in such environments is difficult. Many learning analytics algorithms fail to anticipate consistent, interpretable, and resilient outcomes under real-world operational limitations. The Quantum Residual Attentive Fusion Network (QRAF-Net) is used to provide a neuro-integrated learning outcome prediction framework for session-level categorization in mobile-based remote biology labs. A new Neuro-Integrated Learning Outcome State (NILOS) taxonomy categorizes learner achievement. QRAF-Net captures hierarchical representations and persistent behavioral dependencies via deep residual feature propagation, attention-driven behavioral interaction modeling, and quantum-inspired phase modulation. The system is tested on 254020 real-world laboratory sessions using rigorous preprocessing, dependency-based feature refinement, and hybrid parameter optimization. QRAF-Net outperforms twelve competing deep and transformer-based baseline models by 7--18% in accuracy, precision, recall, and F1-score with 97.8% accuracy. The suggested Outcome Stability Index (OSI) is 0.956, outperforming all baselines in consistency. Ablation experiments, sensitivity assessment, ROC analysis, and resilience testing under network disruptions, behavioral noise, and partial log loss indicate consistent performance decline and good generalization. Explainability findings show that learning gain and procedural consistency dominate, bolstering the framework's clarity and practicality.