FrailNet

Deep survival modeling for rare conflict onset

Survival Analysis
Deep Learning
Conflict Forecasting
Frailnet is a deep survival modeling framework that extends Cox regression to better handle non-proportional temporal dynamics, missingness, and unobserved heterogeneity (frailty).
Published

December 15, 2025

FrailNet is a deep survival modeling framework designed for time-to-event prediction in rare, time-dependent social processes. The model retains a Cox partial-likelihood objective while extending hazard specification to better accommodate non-proportional temporal dynamics, missingness, and unobserved heterogeneity.

Key components include recurrent layers for temporal dependence, an incomplete-aware reconstruction module for denoising and imputation, and multiplicative frailty terms to capture subject-level variation not explained by observed covariates.

Materials

Draft note: This PDF reflects the current draft of an active research manuscript and is being finalized.