Table 1 Comparison of related studies on autism detection from facial static images.
From: A multi-filter deep transfer learning framework for image-based autism spectrum disorder detection
Study (Year) | Model(s) | Classifier(s) | Fine-tuning | Additional techniques |
|---|---|---|---|---|
88 (2021) | VGG-based model | Fully connected layer | Yes | Use of modified pre-trained CNN |
87 (2021) | VGG19 | Fully connected layer | Yes | Transfer learning and dataset evaluation |
79 (2021) | Modified MobileNetV01 | AdaBoost, Decision Tree, Gradient Boost, K-nearest neighbors, Logistic Regression, Multi-Layer Perceptron, Nayve Bayes, Random Forest, SVM, XGB | Both situations | Feature clustering using k-means |
84 (2022) | VGG19, Xception, ResNet50V2, MobileNetV2, EfficientNetB0 | Final Fully-Connected Layers | Yes | Network hyperparameter optimization |
80 (2022) | MobileNet, Xception, EfficientNetB0/B1/B2 | Deep Neural Network (DNN) | Yes | Training loss analysis |
19 (2023) | MobileNet-V2, VGG16 | Logistic Regression, SVM, Random Forest, Decision Tree, Gradient Boosting, Multi-Layer Perceptron, AdaBoost, and K-nearest neighbors | Yes (MobileNetV2) | Preprocessing and normalization over dataset |
91 (2023) | AlexNet, VGG16, VGG19, MobileNet, CNN | Fully-connected layer | Yes | Cloud-edge based structure for educational environments |
90 (2024) | ResNet101, EfficientNetB3 | Self-attention-based Ensemble | Yes | Preprocessing and data augmentation |
Proposed (2025) | ViTSwin, ViT, ViTFER, AffectNet, AlexNet, ResNet-50, VGG16, VGG19 | SVM | No (frozen feature extraction) | Data augmentation, multi-filtering, histogram equalization, dimensionality reduction, scaling normalization |
