A new machine-learning (ML) tool outperforms all existing hepatocellular carcinoma risk prediction methods in patients with confirmed hepatitis C virus eradication after direct-acting antiviral treatment. The study compared five ML tools with current score models in derivation (n = 1,742) and validation (n = 977) cohorts of Japanese adults. SMART, the best-performing ML tool, generates individualized risk-prediction curves using age, body mass index, platelet count and levels of albumin, aspartate aminotransferase, α-fetoprotein and γ-glutamyltransferase.