Backtest Results

Summary Metrics

Total Backtest Years

—

1990 to present · 4 bands

Average ROC-AUC

—

Average across 4 bands

Best Band

—

Highest accuracy

Heidke Skill Score

—

Performance above random

01Band Selector

Eren Bostan·[TAL-601]add per-band backtest chart panels

Band-Level Backtest Charts

Accuracy

—

ROC-AUC

—

HSS

—

Null Model

—

Degradation

Checking...

Yearly Accuracy & ROC-AUC — M4–5

Loading data...

Confusion Matrix (Total)

TP

—

FP

—

FN

—

TN

—

Algorithm Info

Algorithm—

ROC-AUC—

F1 Score—

Accuracy

—

ROC-AUC

—

HSS

—

Null Model

—

Degradation

Checking...

Yearly Accuracy & ROC-AUC — M5–6

Loading data...

Confusion Matrix (Total)

TP

—

FP

—

FN

—

TN

—

Algorithm Info

Algorithm—

ROC-AUC—

F1 Score—

Accuracy

—

ROC-AUC

—

HSS

—

Null Model

—

Degradation

Checking...

Yearly Accuracy & ROC-AUC — M6–7

Loading data...

Confusion Matrix (Total)

TP

—

FP

—

FN

—

TN

—

Algorithm Info

Algorithm—

ROC-AUC—

F1 Score—

Accuracy

—

ROC-AUC

—

HSS

—

Null Model

—

Degradation

Checking...

Yearly Accuracy & ROC-AUC — M7+

Loading data...

Confusion Matrix (Total)

TP

—

FP

—

FN

—

TN

—

Algorithm Info

Algorithm—

ROC-AUC—

F1 Score—

Skill Score

What is the Heidke Skill Score (HSS)?

HSS measures how much better a model performs compared to random guessing. It is the standard reference metric in operational earthquake forecasting systems.

HSS = 2(ad − bc) / [(a+c)(c+d) + (a+b)(b+d)]

a = TP | b = FP | c = FN | d = TN — computed from the full confusion matrix.

HSS < 0Worse than random

HSS = 0Equal to random

HSS > 0Better than random

HSS = 1Perfect

02Year-by-Year Comparison

Defne Yilmaz·[TAL-602]plot all four bands yearly accuracy

4 Bands — Yearly Accuracy

Preparing chart...

03Model History

Eren Bostan·[TAL-603]render model retraining timeline

Retraining Timeline

Loading retraining history...

Scheduled Degradation Online Learning

Backtest runs automatically every Monday. The latest results will be displayed above.