Table 2 Comparative analysis of prior models and identification of research gaps.
Study | Core Model Architecture | Handles Sequential/Temporal Dependencies? | Includes Feature Attention/Weighting? | Parameter Count (Approx.) | Identified Gap |
|---|---|---|---|---|---|
Akter et al. | MobileNetV1 | No | No | 4.2M | Lacks focus on feature importance and sequential relationships. |
Alam et al. | Xception & ResNet50 | No | No | 25.6M | Standard architecture; does not prioritize specific facial regions. |
Cao et al. | ViT | Yes (Implicitly) | Yes (Self-Attention) | - | While powerful, ViT can be data-hungry and computationally intensive. |
Rabbi et al. | VGG19 & InceptionV3 | No | No | 143.7M | Relies on conventional feature extraction without specialized focus. |
Mujeeb Rahman et al. | Xception & EfficientNet | No | No | - | Focuses on architectural comparison, not on modeling feature dependencies. |
Our Proposed Model | Multi-Residual + BiLSTM + Attention | Yes (BiLSTM) | Yes (Attention Layer) | 28.5M | Addresses all identified gaps in a unified framework with moderate efficiency. |
