KAN-AFT: Interpretable Nonlinear Survival Model with Kolmogorov-Arnold Networks and Accelerated Failure Time Analysis

This ArXiv paper introduces KAN-AFT, a novel survival analysis model that combines Kolmogorov-Arnold Networks (KANs) with Accelerated Failure Time (AFT) analysis. The key innovation lies in addressing the interpretability limitations of deep learning models like DeepAFT, while maintaining comparable or superior performance. By leveraging KANs, the model can represent complex nonlinear relationships and provide symbolic equations for survival time, enhancing understanding of the model's predictions. The paper highlights the AFT-KAN formulation, optimization strategies for censored data, and the interpretability pipeline as key contributions. The empirical results suggest a promising advancement in survival analysis, balancing predictive power with model transparency. This research could significantly impact fields requiring interpretable survival models, such as medicine and finance.

• •Introduces KAN-AFT, a novel interpretable survival analysis model.
• •Combines Kolmogorov-Arnold Networks with Accelerated Failure Time analysis.
• •Addresses the interpretability limitations of deep learning models in survival analysis.