Table 3 Performance comparison of building damage assessment models for hurricane-induced damage detection for Hurricane Ian case study in Lee County, Florida

From: Scalable variational learning for noisy-OR Bayesian networks with normalizing flows for complex cascading disaster systems

Model

TPR

TNR

AUC

Online-DisasterVINF*

0.8173

0.7721

0.7883

Bayesian network-based Model*27

0.8293

0.6221

0.7553

FCS-Net (w/o finetuning)48

0.2713

0.8941

–

FCS-Net (w/ finetuning)

0.2098

0.9386

–

Dual-HRNet (w/o finetuning)49

0.0912

0.9795

–

Dual-HRNet (w/ finetuning)

0.8217

0.6251

–

DPM-based Model*

0.6498

0.6249

0.6739

Fragility Curve*

0.5669

0.6246

0.5695

  1. *Models trained without labels.
  2. TPR true positive rate, TNR true negative rate, AUC area under curve,DPM damage proxy map, w/ with, w/o without.
  3. Finetuning: Model adaptation using Hurricane Ian labeled data.
  4. ‘–’ indicates metric not available for deterministic models.
Back to article page