Idea
Interpretable physics-informed ML model for accurate earthquake shaking prediction benefiting engineers and disaster planners.
Research Paper
Core Innovation
This paper introduces a physics-informed ML model that inherently interprets input contributions while addressing data imbalance with HazBinLoss. Unlike traditional black-box models, it prioritizes critical near-fault events for better damage prediction. The approach matches established ground motion model performance with improved transparency.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: global seismic risk assessment and structural engineering markets require advanced predictive models.
Potential Customers & Pain Points
- Structural engineers needing reliable seismic risk data
- Disaster planning agencies requiring transparent earthquake models
- Insurance companies assessing earthquake damage risk
- Urban developers seeking accurate ground motion forecasts
- Researchers addressing imbalanced seismic datasets
Business Model
Licensing the ML model as an API to engineering firms and disaster agencies; consulting for model integration and customization.
Competitive Landscape
- OpenQuake
- ShakeMap
- DeepShake
Implementation Challenges
- Adoption resistance due to trust in traditional models
- Data availability and quality for near-fault events
- Integration with existing seismic risk workflows
Validation Strategy
- Benchmark model against established GMMs on diverse seismic datasets
- Pilot deployment with structural engineering firms for real-world feedback
- Collaborate with disaster agencies to validate interpretability and usability
Research Paper Overview
Breaking the Black Box: Inherently Interpretable Physics-Informed Machine Learning for Imbalanced Seismic Data
Summary
Ground motion models predict earthquake shaking intensity and are vital for structural analysis and seismic risk assessment. Traditional ML models lack interpretability and struggle with imbalanced data favoring less damaging distant records over critical near-fault events. This work introduces a transparent ML architecture using HazBinLoss to weight critical near-field large magnitude records higher, ensuring accurate prediction of damaging scenarios. The model aligns with seismological principles and matches established GMM performance while providing clear input contributions, enabling trust and broader adoption in disaster planning.