Transformer-Based Survival Analysis for Predicting Policyholder Retention in Life Insurance

DOI

10.2991/978-94-6463-992-6_30How to use a DOI?

Keywords

Survival Analysis; Transformer Model; Life Insurance; Deep Learning; Actuarial Risk Prediction

Abstract

Survival analysis plays a pivotal role in the insurance sector, particularly for modeling time-to-event data such as policy lapses or withdrawals, which directly influence actuarial pricing, reserve estimation, and risk management. Traditional methods like the Cox Proportional Hazards model are limited in handling non-linear relationships and time-varying covariates. This paper proposes a Transformer-based framework, adapted from recent deep learning advancements, to improve prediction accuracy for policy-holder retention in life insurance. The model employs self-attention mechanisms to capture complex feature interactions, including age, gender, and premium payments, while effectively managing censored observations. A small innovation involves an insurance-specific embedding layer to prioritize domain-relevant covariates. Experiments on a synthetic dataset demonstrate the model’s superiority, achieving a C-index of 0.81, Integrated Brier Score (IBS) of 0.09, and log-likelihood of -0.78, outperforming baselines like CoxPH (C-index 0.49, IBS 0.14) and DeepSurv (C-index 0.03, IBS 0.11). Visualizations, such as Kaplan-Meier curves, show gender-stratified survival probabilities with females retaining longer. Attention heatmaps reveal key feature weights, emphasizing age. Training curves indicate stable convergence, cumulative incidence plots illustrate competing risks, and calibration plots confirm reliability across risk groups. Ablation studies validate the attention and embedding components, improving performance by 10–12%. This approach not only enhances predictive power but also offers interpretable insights for insurers to mitigate churn and optimize strategies. Future work could integrate real-time data for dynamic updates. To enhance real-world applicability, future studies should validate the model with actual insurance data, addressing potential irregularities in policyholder behavior.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Dongsheng Lyu
PY  - 2026
DA  - 2026/02/20
TI  - Transformer-Based Survival Analysis for Predicting Policyholder Retention in Life Insurance
BT  - Proceedings of the 2025 4th International Conference on Mathematical Statistics and Economic Analysis (MSEA 2025)
PB  - Atlantis Press
SP  - 319
EP  - 327
SN  - 2352-5428
UR  - https://doi.org/10.2991/978-94-6463-992-6_30
DO  - 10.2991/978-94-6463-992-6_30
ID  - Lyu2026
ER  -