Interpreting machine learning models for survival analysis: a study of cutaneous melanoma using the SEER database

In this study, we train and compare three types of machine learning algorithms for Survival Analysis: Random Survival Forest, DeepSurv and DeepHit, using the SEER database to model cutaneous malignant melanoma. Additionally, we employ SurvLIMEpy library, a Python package designed to provide explainability for survival machine learning models, to analyse feature importance. The results demonstrate that machine learning algorithms outperform the Cox Proportional Hazards Model. Our work underscores the importance of explainability methods for interpreting black-box models and provides insights into important features related to melanoma prognosis.
This research was supported by the Spanish Research Agency (AEI) under projects PID2020-116294GB-I00 and PID2020-116907RB-I00 of the call MCIN/ AEI /10.13039/501100011033, the project 718/C/2019 with id 201923-31 funded by Fundació la Marato de TV3 and the grant 2020 FI SDUR 306 funded by AGAUR.
Peer Reviewed
Postprint (author's final draft)