<![CDATA[Anemia is a global health problem affecting approximately 1.2 billion people worldwide, with the highest prevalence among pregnant women and adolescent girls. Conventional diagnosis through laboratory blood tests is invasive, requires trained medical personnel, and is unaffordable for communities in remote areas. This study aims to evaluate and compare the performance of three deep learning architectures, namely DenseNet121, EfficientNetB0, and ResNet-50, in classifying anemia and non-anemia conditions non-invasively based on palm images. The dataset used is a public dataset called anemiatangan from the Kaggle platform, consisting of 10,200 images with two classes, Anemia and Non-Anemia, divided into 80% training data (8,200 images), 10% validation (1,000 images), and 10% testing (1,000 images). All three models were trained using a transfer learning approach with pre-trained weights from ImageNet, accompanied by preprocessing and data augmentation. Evaluation was performed based on accuracy, precision, recall, F1-Score, and AUC-ROC (Area Under the Receiver Operating Characteristic Curve) metrics. The test results indicate that DenseNet121 and EfficientNetB0 achieved the highest accuracy of 99% with precision, recall, and F1-Score values approaching perfection at 0.99, while ResNet-50 recorded an accuracy of 97%. Therefore, DenseNet121 and EfficientNetB0 are proven to be the most optimal architectures for implementing a non-invasive anemia screening system based on palm images, with the potential to be integrated into mobile applications and telemedicine systems to support early detection of anemia in remote areas.]]>
Copyrights © 2026
