Early cancer detection remains a critical challenge in both Japan and Malaysia, where rising incidence rates and frequent diagnoses at advanced stages limit treatment options and worsen patient outcomes. To address this, we aim to develop a minimally invasive blood-based diagnostic system using circulating RNA molecules in blood, known as cell-free RNA (cfRNA). cfRNA carries distinct expression signatures that reflect early tumor development, making it a promising candidate for early detection. In this study, we aim to establish a scalable and accurate cfRNA-based diagnostic framework. We profiled millions of cfRNA molecules, followed by biomarker discovery using Forward Greedy Search (FGS) and Lasso regularization. The selected RNA candidate features were then evaluated using prediction models built with Random Forest (RF) and other machine learning algorithms. We systematically benchmarked components of the pipeline, including input structure, biomarker set size, and modeling strategy, to identify the most effective configuration for liver cancer diagnosis.