<![CDATA[Anemia is a growing global health concern, driving the need for non-invasive detection methods. This research develops a non-invasive hemoglobin (Hb) level prediction model utilizing Photoplethysmography (PPG) signals and the Extreme Gradient Boosting (XGBoost) algorithm, addressing the limitations of conventional invasive and time-consuming approaches. PPG signals, captured by optical sensors (red 660 nm and infrared 880 nm) on the fingertip, monitor blood volume changes that correlate with Hb levels based on the Beer-Lambert Law. Following pre-processing of secondary data from 68 subjects (including missing value handling and gender encoding) and average Systolic Peak feature extraction, the XGBoost model was trained and evaluated. To enhance performance and overcome data limitations, data augmentation was implemented, expanding the sample to 204. Evaluation results demonstrate significant improvement: on the original data, the model achieved an MAE 0,0769, RMSE 0,1117, and R² 0,4954. For the post-augmentation, performance drastically improved to an MAE 0,0190, RMSE 0,0254, and the R² of 0,9724. This increased R² indicates the model's ability to capture 97,24%; of hemoglobin variability, while reduced MAE and RMSE signify higher prediction accuracy and better generalization, making this model reliable for non-invasive Hb prescreening and potentially supporting anemia diagnosis.]]>
Copyrights © 2025
