<![CDATA[Monitoring hemoglobin (Hb) concentration is essential for the early detection and management of anemia, yet conventional assessment relies on invasive blood sampling and laboratory facilities, limiting its suitability for routine screening. This study developed and validated a portable non-invasive system for hemoglobin estimation and real-time anemia classification by integrating a MAX30102 optical sensor, Raspberry Pi, and a K-Nearest Neighbors (KNN) machine learning algorithm. Photoplethysmography (PPG) signals acquired from fingertip measurements were processed to extract representative optical features, while invasive hemoglobin measurements obtained using a calibrated commercial analyzer served as the reference standard. Experimental validation demonstrated that the proposed KNN model accurately estimated hemoglobin concentration and consistently outperformed a conventional linear regression approach in capturing the nonlinear relationship between PPG features and hemoglobin levels. The system successfully classified users into normal, mild, and moderate anemia categories and displayed the results in real time through both OLED and web-based interfaces. Although no severe anemia cases were included in the validation dataset, the prototype showed reliable performance within the evaluated hemoglobin range. These findings demonstrate the feasibility of combining optical sensing and machine learning for portable, non-invasive hemoglobin monitoring. Further validation involving larger and more clinically diverse populations is required before clinical implementation.]]>
