A Neural Network Optimization Approach for Anemia Prediction Using Image Pixels and Hemoglobin Data

This study developed a neural network optimization model to predict anemia using image pixel data and hemoglobin levels. The model was designed to address class imbalance and achieved strong performance, with a test accuracy of 97%, precision of 97%, and recall of 96.88%. Data preprocessing techniques such as z-score normalization and ordinal encoding were applied to standardize the input features, while L2 regularization was used to prevent overfitting. The model's performance was assessed using multiple metrics, such as the F1-score and AUC, highlighting its reliability in anemia prediction. The results indicate that this model offers a reliable and efficient approach for non-invasive anemia detection, making it a valuable tool for clinical use in diagnosing anemia based on image data and hemoglobin levels.