Abstract
The use of deep learning methods has gained momentum in the domain of survival analysis. Different models have been proposed to handle time-to-event data. Neural networks are used to find complex relationships between features, improving the predictive capabilities of deep learning models. When conducting experiments, one might want to reduce the number of methods that need to be examined because of the computational resources required for model training. Establishing families of deep learning methods that behave in a similar way might be beneficial for such scenarios. In this paper, we establish a way to measure differences between deep learning discrete-time survival analysis models. The proposed method is based on SHAP values. We conducted experiments for three datasets and five discrete-time survival analysis models. We proposed a special kind of plot that helps visualize the impact of features on the model outputs over time intervals. Based on the obtained results, we performed Friedman and Wilcoxon tests to examine statistically significant differences between the models.
Paper Type
Full Paper
DOI
10.62036/ISD.2025.53
Finding differences between discrete-time deep learning survival models
The use of deep learning methods has gained momentum in the domain of survival analysis. Different models have been proposed to handle time-to-event data. Neural networks are used to find complex relationships between features, improving the predictive capabilities of deep learning models. When conducting experiments, one might want to reduce the number of methods that need to be examined because of the computational resources required for model training. Establishing families of deep learning methods that behave in a similar way might be beneficial for such scenarios. In this paper, we establish a way to measure differences between deep learning discrete-time survival analysis models. The proposed method is based on SHAP values. We conducted experiments for three datasets and five discrete-time survival analysis models. We proposed a special kind of plot that helps visualize the impact of features on the model outputs over time intervals. Based on the obtained results, we performed Friedman and Wilcoxon tests to examine statistically significant differences between the models.
Recommended Citation
Magnuszewski, P. & Kretowska, M. (2025). Finding differences between discrete-time deep learning survival modelsIn I. Luković, S. Bjeladinović, B. Delibašić, D. Barać, N. Iivari, E. Insfran, M. Lang, H. Linger, & C. Schneider (Eds.), Empowering the Interdisciplinary Role of ISD in Addressing Contemporary Issues in Digital Transformation: How Data Science and Generative AI Contributes to ISD (ISD2025 Proceedings). Belgrade, Serbia: University of Gdańsk, Department of Business Informatics & University of Belgrade, Faculty of Organizational Sciences. ISBN: 978-83-972632-1-5. https://doi.org/10.62036/ISD.2025.53