Weather‑Damage Forecasting

Predicted damage using LSTM‑based time‑series models; compared ARIMA, LSTM, and Transformer approaches.

January 1, 2025

Python XGBoost PyTorch Predictive Analysis Data Visualization Data Analysis LSTM Machine Learning

Problem

Forecast weather‑driven damage risk to support insurance pricing and mitigation planning.

Approach

Engineered features from historical weather and claims data; handled seasonality and missing records.
Trained LSTMs and compared against ARIMA and Transformer baselines with sliding‑window evaluation.
Built visual dashboards for error analysis and scenario exploration.

Weather Damage Visualization

Impact

Achieved >98% accuracy on target labels in validation tests.
Informed pricing strategies and resource allocation under forecast uncertainty.

Tech

Time‑series: supervised windows, cross‑validation
Models: LSTM, ARIMA, Transformer; XGBoost for tabular blends