<![CDATA[Anemia is a global health problem that often goes undetected early because conventional diagnostic methods are invasive, require specialized laboratory equipment, and are relatively expensive. Clinically, pallor in the palpebral conjunctiva of the eye is a primary physical indicator of hemoglobin deficiency. This study aims to develop a Deep Learning-based, non-invasive anemia detection system using a Convolutional Neural Network (CNN) algorithm with the EfficientNet-B0 architecture. This model was selected due to its high computational efficiency and minimal number of parameters (5.3 million), making it suitable for implementation in resource-constrained environments. The dataset used consists of 670 images of the palpebral conjunctiva obtained from the Mendeley Data repository. To overcome the limitations of the dataset size, a Two-Stage Training strategy (Frozen Warm-Up and Unfrozen Fine-Tuning phases) along with dynamic data augmentation techniques were applied. Evaluation results on independent test data demonstrated that the model achieved an Accuracy of 82%, with a Precision of 85%, a Sensitivity (Recall) of 80%, and an F1-Score of 82%. The high sensitivity value indicates the model's reliable capability in identifying positive cases of anemia. Error analysis revealed that prediction failures were predominantly caused by low lighting factors (underexposure). In conclusion, the proposed method is proven to be effective and holds the potential to be used as a fast, accurate, and non-invasive early screening tool for anemia.]]>
