Sistem Cerdas Deteksi Risiko Anemia Berbasis Hybrid Convolutional Neural Network dan Expert System pada Analisis Citra Konjungtiva Mata dan Gejala Klinis Pasien

Abstrak

Anemia remains a global health problem, requiring early detection to prevent serious complications. Hemoglobin testing is invasive, while previous non-invasive screening approaches rely on a single parameter, limiting early detection effectiveness. This study develops a non-invasive anemia screening system using conjunctival images and clinical symptoms based on a Convolutional Neural Network (CNN) with MobileNetV2, an expert system, and a weighted hybrid method. A total of 3,870 conjunctival images were used for training and validation, while 50 test samples were collected using a smartphone and clinical symptom data. The results show that the CNN achieved 94% accuracy, the expert system 90%, and the hybrid method achieved 96% accuracy, 100% precision, 89% recall, and a 94% F1-score. These findings indicate that the integration of methods improves screening performance and supports a fast, easy, non-invasive, and practical anemia screening system for early detection that can be used independently at home with accessible devices.

Identifier

• DOI: https://doi.org/10.29240/arcitech.v6i1.17146
• OpenAlex: https://openalex.org/W7167711909