Improved Anemia Medical Diagnosis on Complete Blood Count: Tuning Projected Long-Short Term Memory Layers with Coefficient of Determination

Anemia, characterized by a deficiency in red blood cells (RBCs) or hemoglobin (HGB), poses significant global public health challenges. Accurate diagnosis and classification of anemia types, crucially dependent on hemoglobin (HGB) level estimation, is essential for effective management and treatment planning. This study investigates the prevalence, severity, and demographic associations of various anemia types among 364 adult patients using publicly available data from complete blood count (CBC) tests. The dataset includes 11 CBC attributes pivotal for anemia diagnosis, encompassing HGB levels and RBC counts. To address the complexities inherent in CBC data, such as variations in normal ranges based on age and gender, and challenges in data interpretation across diverse patient demographics, machine learning techniques are employed. Specifically, projected layers of long short-term memory networks (PLSTM) are utilized. These techniques require robust methodologies for data preprocessing, feature selection, and model optimization based on coefficient of determination R2, prioritizing realism and the preservation of critical medical information. This approach ensures accurate predictive modeling of anemia types based on hemoglobin (HGB) levels, classified as mild, moderate, or severe according to World Health Organization (WHO) guidelines. Insights derived from this study, leveraging PLSTM over traditional LSTM networks, contribute to enhancing diagnostic accuracy and developing personalized treatment strategies for anemia. This underscores the role of machine learning in tailoring approaches to complex medical datasets, thereby advancing healthcare outcomes.