WEDNESDAY, Dec. 3, 2025 (HealthDay News) -- The CatBoost model has the best overall performance and stability among different machine learning algorithms for predicting survival rates for patients with melanoma, according to a study published online Nov. 28 in Scientific Reports.

Enbo Hu, from Shenzhen People's Hospital in China, and colleagues applied five machine learning models, incorporating 13 demographic and clinical variables, to predict survival rates for patients with melanoma in an effort to improve prognostic accuracy and support clinical decision-making. Data were analyzed for 4,875 patients with cutaneous melanoma.

The researchers found that following fivefold cross-validation, the CatBoost model demonstrated the best overall performance and stability. For one-, three-, and five-year survival predictions, the model achieved area under the curve values of 0.7577, 0.7595, and 0.7557, respectively. Clinical utility was further confirmed in a decision curve analysis; across training and test sets, consistent precision indicated robust generalization and reliable predictive capability.

"This study establishes a robust and interpretable machine learning-based model for melanoma survival prediction, demonstrating that CatBoost effectively balances accuracy, generalizability, and clinical relevance," the authors write.

Abstract/Full Text