Hepatitis B virus (HBV) is a major risk factor for hepatocellular carcinoma (HCC), characterized by genomic instability and chronic DNA replication stress. This study presents a robust machine-learning framework using random survival forest to develop a DNA replication stress-related prognostic index (PI_RS) for HBV-associated HCC. Transcriptomic expression profiles from 606 HCC cases were used to construct PI_RS, which outperformed population-based predictors, demonstrating superior prognostic prediction in HBV-associated HCC. Lower PI_RS scores were associated with higher expression of HBV oncoproteins, activated immune/metabolism pathways, and increased responsiveness to immunotherapy. Conversely, higher PI_RS scores correlated with elevated Ki-67 marker, cancer stemness, and enrichment in DNA replication stress, cell cycle pathways, and chromatin remodelers, resulting in an 'immune-cold' phenotype and unfavorable clinical outcomes. Through large-scale in-silico drug screening, potential therapeutic targets (TOP2A, PRMT1, CSNK1D, and PPIH) and five agents, including topoisomerase and CDK inhibitors, were identified for patients with high PI_RS scores. These findings hold promise for optimizing therapeutic strategies in HCC and providing insights into the management of HBV carriers. In summary, our machine-learning approach yielded PI_RS as a powerful predictor for assessing prognosis in HBV-associated HCC. This analytic framework improves population-based therapeutic strategies, facilitates personalized treatment, and ushers in a new era of precision medicine in HCC.