<![CDATA[This study aims to develop a Deep Learning model based on the Feature Tokenizer Transformer (FTTransformer) architecture for blood disease classification using 103,024 retrospective secondary Complete Blood Count (CBC) patient data from Royal Prima Hospital Medan. The self-attention mechanism in this model is implemented to automatically map complex interactions among hematological parameters without requiring manual feature engineering. Testing results demonstrate that the FT-Transformer effectively overcomes the challenges of highly imbalanced clinical data, yielding superior and stable multi-class classification performance. This is evidenced by Accuracy, Precision, Recall, and F1-Score metrics reaching 0.98 to 1.00 across four main diagnostic categories: Normal, Anemia, Sepsis/Infection, and Thrombocytopenia. Overall, this computational approach successfully produced a robust and high-precision Clinical Decision Support System (CDSS) prototype for interpreting tabular laboratory results.]]>
Copyrights © 2026
