Rek-Surv: A lightweight deep survival model for plant infectious disease onset prediction

Area of Science:

• Agricultural Science
• Computational Biology
• Epidemiology

Background:

• Infectious crop diseases threaten global food security, necessitating timely detection and intervention.
• Traditional survival analysis models struggle with high-dimensional agricultural data, leading to poor performance and interpretability.
• Existing deep survival models like MLPs face challenges with overfitting and parameter efficiency in complex datasets.

Purpose of the Study:

• To introduce a novel deep survival model for predicting the timing of infectious disease onset in both agricultural and clinical settings.
• To leverage the Kolmogorov-Arnold Network (KAN) architecture for improved predictive accuracy and computational efficiency in survival analysis.
• To develop a lightweight and interpretable deep survival model suitable for real-time outbreak detection.

Main Methods:

• Developed Rek-Surv, a deep survival model utilizing an Efficient-KAN backbone with residual connections and enhanced regularization.
• Applied survival analysis techniques adapted for plant disease surveillance and human epidemiology.
• Evaluated Rek-Surv on five clinical benchmark datasets and a citrus Huanglongbing (HLB) plant disease dataset.

Main Results:

• Rek-Surv achieved a high concordance index (C-index) of 0.962 on the HLB dataset with only 114 trainable parameters.
• Demonstrated millisecond-level inference speed, indicating high computational efficiency.
• Outperformed existing survival models in accuracy and complexity, showing generalizability across human and plant infectious diseases.

Conclusions:

• Rek-Surv offers a highly efficient and accurate solution for predicting infectious disease outbreak timing.
• The model's lightweight architecture and strong performance make it ideal for real-time outbreak detection and proactive disease management.
• Advanced survival models like Rek-Surv can significantly enhance infectious disease control strategies in agriculture and healthcare.