Hurricane-Induced-Economic-Loss-Prediction

This work appears in SpatialConnect‘25: Proceedings of the 1st ACM SIGSPATIAL International Workshop on Spatial Intelligence for Smart and Connected Communities.

How to Cite

If you find this work useful in your research, please consider citing:

@inproceedings{10.1145/3764924.3770889,
author = {Shen, Bolin and Ozguven, Eren and Zhao, Yue and Wang, Guang and Xie, Yiqun and Dong, Yushun},
title = {Learning from the Storm: A Multivariate Machine Learning Approach to Predicting Hurricane-Induced Economic Losses},
year = {2025},
url = {https://doi.org/10.1145/3764924.3770889},
location = {The Graduate Hotel Minneapolis, Minneapolis, MN, USA},
}

Install Environment

We use uv to manage this project.

uv sync

Packages we use

hurricane v0.1.0
├── geopandas v1.0.1
├── matplotlib v3.9.4
├── pandas v2.2.3 (*)
├── scikit-learn v1.6.1
├── scipy v1.13.1 (*)
├── shapely v2.0.7 (*)
└── xgboost v2.1.4

Run Experiments

Predict economic loss.

model_name: Name of the model to run (e.g., RF, XGB, NN, GBM, Stacked)

python main.py --model_name XGB

Example:

Evaluating RF with 5-fold cross-validation:
  Fold 1 metrics: R2: 0.7264, MAE: 1.1896, SMAPE: 6.4793, RMSE: 1.6575, RMSLE: 0.0824
  Fold 2 metrics: R2: 0.6310, MAE: 1.4707, SMAPE: 7.9966, RMSE: 1.8900, RMSLE: 0.0957
  Fold 3 metrics: R2: 0.7012, MAE: 1.2913, SMAPE: 7.4189, RMSE: 1.7228, RMSLE: 0.0915
  Fold 4 metrics: R2: 0.6522, MAE: 1.3366, SMAPE: 7.3960, RMSE: 1.8208, RMSLE: 0.0948
  Fold 5 metrics: R2: 0.6633, MAE: 1.4156, SMAPE: 7.8946, RMSE: 1.8395, RMSLE: 0.0930

RF performance (5-fold CV):
R2: 0.675 \pm 0.0
MAE: 1.341 \pm 0.1
SMAPE: 7.437 \pm 0.5
RMSE: 1.786 \pm 0.1
RMSLE: 0.091 \pm 0.0

Visualization

plot importance

python visualization.py

plot heatmap

python visualization.py