Facial Image-Based Autism Detection Using A CNN-ExtractedFeature Ensemble with Random Forest Classification

For effective care and attention, autism spectrum disorder (ASD) must be identified early and ‎This study suggests a hybrid model that blends convolutional neural networks (CNN) and ‎random forest (RF) classifiers. The proposed model had a high classification accuracy of ‎‎91.3%, demonstrating its ability to reliably detect cases of ASD. With a precision of 90.7%, ‎the model effectively reduces false positives, and its 89.5% recall ensures excellent sensitivity ‎to actual ASD patients. The balanced F1-Score of 90.1% shows that the model performs ‎consistently in terms of both precision and recall. Additionally, the model's Area Under the ‎Curve (AUC) score of 0.94 indicates its exceptional ability to discriminate between ASD and ‎non-ASD classes. All of these results demonstrate the Hybrid CNN + RF model's strength ‎and dependability, which makes it a promising non-invasive technique for early facial feature-based ASD screening‎.