Machine learning-based model for predicting malignant melanoma survivability

LIN Chong; DONG Bin; DAI Zhihui; ZHANG Lulu; WU Lei

doi:10.3969/j.issn.1673-7040.2025.04.007

Abstract：Objective Based on different machine learning algorithms,a prognostic prediction model is established to guide clin-icians to screen high-risk patients and further improve patient survival by taking corresponding treatment measures.Methods Collecting the data of patients diagnosed with cutaneous malignant melanoma in the SEER database from 2004 to 2013,using logistic regression,de-cision tree,naïve Bayes,K-nearest neighbor and random forest algorithm in machine learning,a model was established to predict the 5-year survival status of cutaneous malignant melanoma,and the 10-fold cross-validation method was used to verify the model effective-ness and evaluate the prediction efficacy of different models.Results We include a total of 92 328 participants,including the training set(n=64 630,70％)and the test set(n=27 698,30％);5-year survival rate for all patients is 92.5％.In this study,the area under the receiver operating characteristic curve of logistic regression,decision tree,naïve Bayesian,K-neighbor,and random forest models were 0.908,0.858,0.907,0.724,and 0.903.Conclusion This study predicted the survival status of patients with cutaneous malignant melanoma 5 years later based on machine learning algorithms.Logistic regression,naive Bayes and random forest models were more effective in predicting the survival of malignant melanoma,providing a prediction tool different from traditional statistical methods.It can assist clinical physicians in screening high-risk patients and provide precise individualized treatment and personalized prognosis management for patients with malig-nant melanoma.