97 Machine learning algorithms in the prognosis of cutaneous melanoma: a population-based study.

Abstract

Objectives: To establish a predictive model for prognosis of cutaneous melanoma using machine learning algorithms in large sample data.

Methods: A retrospective analysis of patients diagnosed with cutaneous melanoma in the SEER database from 2010 to 2015 was performed using 12 different machine learning algorithms, for a total of 97 algorithm combinations, to screen for variables associated with cutaneous melanoma prognosis and to build predictive models.

Results: A total of 24,457 cases were collected in this study, and 8,441 cases were finally included. Among them, 5908 cases in the training set and 2533 cases in the test set. The results of the study show that StepCox[both] + RSF is the best model. The variable features screened by the best model were Sex, Age, Marital, T stage, N stage, Ulcer, Site, Histologic, Surgery, Chemotherapy, Bone metastasis, Liver metastasis and Lung metastasis.

Conclusion: We have developed a predictive model with good accuracy for cutaneous melanoma prognosis using a combination of 97 machine learning algorithms in a large sample database.