Mortality Prediction Model for Life Insurance
BUSINESS PROBLEM
- A leading life insurance company in India, with products across term, non-term, and mixed segments, maintained a sub-1% claim rate for its term portfolio.
- To improve underwriting and claims management, there was a need to build an early claims forecasting system using multi-source data—enabling proactive risk assessment and faster decision-making across all products.
SOLUTION
- Multi-Stage ML Framework to enhance decision-making across the customer journey.
Stage 1: Predict claim risk using application profile data.
Stage 2: Refined predictions using enriched data—profile, IIB, and medical information. - Void Classification Model developed to minimize future manual voids by automating identification.
- Segmented models built to address loss prediction across 1-year vs. multi-year policy durations.
- Hyperparameter optimization through iterative experimentation with diverse ML algorithms and frameworks for optimal model performance.
BENEFITS
- Risk Segmentation of Applications in the Customer Journey.
- Optimization of Due Diligence Efforts.
- Enabling instruments of differential pricing based on data and history.
- Automated underwiring to minimize human intervention, easy to scale on demand.
PERFORMANCE
Model Results
MODEL NAME | CAPTURE RATE IN TOP 5 PERCENTILE |
TERM Product 1 YEAR CLAIM | 39.53% |
TERM Product 1 YEAR VOID | 60.96% |
HYBRID Product 1 YEAR CLAIM | 42.86% |
HYBRID Product 1 YEAR VOID | 49.23% |
TERM Product 3 YEAR CLAIM | 34.63% |
Performance measurement of Model at various thresholds - ROC curve
