Detection of Neonatal Jaundice with Hybrid Models of ViT and Machine Learning ViT ve Makine Öǧrenmesi Hibrit Modelleri ile Yenidoǧan Sariliǧi Tespiti
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636687
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Deep Feature Extraction, Machine Learning Classification, Neonatal Jaundice, Vision Transformer
- Trakya Üniversitesi Adresli: Evet
Özet
This study proposes a hybrid approach combining Vision Transformer (ViT)-based deep feature extraction with classical machine learning classifiers for the detection of neonatal jaundice in images. The high-dimensional feature vector obtained from the developed ViT model was classified using SVM, MLP, Logistic Regression, KNN, RF, XGBoost, Naive Bayes, and SVM-MLP-based Soft-Ensemble models, and the model performances were comparatively analyzed. According to the results, it was observed that the MLP and Ensemble models, in particular, offered more balanced and reliable performance compared to the others in terms of accuracy, F1, and AUC metrics. These results demonstrate that hybrid approaches combining deep feature extraction with classical classifiers can be used as an effective jaundice detection decision support tool in small datasets.