REK-SURV: A High-Efficiency Deep Survival Analysis Model Based on Kolmogorov-Arnold Networks with a Residual Mechanism

Agricultural disease prediction and early warning systems are essential for safeguarding crop health, optimizing yield, and ensuring global food security. These systems rely on robust statistical methods for accurate disease outbreak forecasting. Survival analysis, a widely used approach in medical research, presents a promising solution due to its ability to model time-dependent events and predict critical occurrences, such as the onset of diseases in crops. In deep survival analysis, Multilayer Perceptrons (MLPs) are widely used models; however, they face significant limitations when applied to high-dimensional data, including overfitting, lack of interpretability, low parameter efficiency, and processing bottlenecks. To address these challenges, we introduce the Kolmogorov-Arnold Network (KAN) in the context of deep survival analysis. KAN introduces an innovative network architecture that effectively mitigates the aforementioned issues, while demonstrating exceptional function approximation capabilities and enhanced parameter efficiency. We propose REK-SURV, a deep survival model based on the KAN architecture, which incorporates a residual mechanism and an enhanced loss function. We evaluate the model using five real-world medical datasets, and the results demonstrate that REK-SURV significantly outperforms existing models, achieving superior c-index values with considerably fewer parameters and faster inference speed. By leveraging the highly efficient 'Efficient KAN' architecture, REK-SURV not only accelerates both training and inference processes but also offers a distinct advantage for deployment in resource-constrained environments.