Table 1 Comparison of the existing methods
Ref | Authors (Year) | Methodology / Focus | Strengths | Limitations |
|---|---|---|---|---|
17 | Al-Sharafi et al. (2023) | Hybrid SEM-ANN to study sustainable AI chatbot use in education | Strong theoretical modeling highlights the importance of institutional factors | No real-time adaptation or behavior monitoring |
18 | Aung et al. (2024) | Docker-based Flutter learning environment | Improves learning accessibility and consistency | Lacks behavioral or engagement tracking |
19 | Han et al. (2025) | WAD-YOLOv8 for classroom student behavior detection | Accurate visual recognition of discrete student actions | Ignores temporal and emotional behavior context; high computational cost |
20 | Sheng et al. (2025) | Optimized YOLOv8s for real-time behavior detection | Improved detection speed and accuracy in dense settings | No multimodal fusion or QoS adaptation |
21 | Johnson et al. (2024) | Blended learning with a single-case design | Improves student performance and engagement | No AI integration; limited scalability |
22 | Shiri et al. (2024) | EfficientNetV2-L + RNN for engagement detection | Captures temporal engagement patterns; high accuracy | High processing demand; no real-time orchestration |
23 | Fazil et al. (2024) | Deep learning for student performance prediction | Strong predictive ability; uses engagement features | Post-hoc analysis only; no live adaptation |
24 | Li et al. (2024) | Multimodal graph learning with 3D Haar framelets | High-order modality interaction modeling | Extremely resource-intensive; unsuitable for edge devices |
25 | Chong et al. (2024) | Belief Rule-Based (BRB) model for engagement level categorization | Interpretable, expert-informed framework | Static rules; lacks adaptive learning capabilities |
26 | Prameela et al. (2024) | Multi-head neural network for behavior prediction + feedback | Allows dynamic behavior classification; real-time feedback | No resource-aware adaptation; needs stable environments |
27 | Maddu & Murugappan (2024) | Facial emotion recognition using a hybrid classifier | Effective in emotion detection during online learning | Privacy issues; requires continuous video input |
28 | Fan & Tian (2024) | fsQCA analysis on online learning satisfaction | Reveals complex combinations affecting learner satisfaction | No AI automation; lacks real-time inference |
29 | Somu & Kumar (2024) | Emotion-based engagement modeling with Adam optimizer | Emotion-driven personalization using robust optimization | Batch processing; no edge deployment or QoS control |