Model Accuracy & Backtesting

Overall Performance (2017-2025)

82.3%
Accuracy
79.1%
Precision
85.6%
Recall
0.89
AUC-ROC

How We Measure Success

  • • True Positive: Model predicts layoff ≥75%, layoff occurs within 1-3 months
  • • True Negative: Model predicts <25%, no layoff occurs
  • • False Positive: High prediction, no layoff (cost: unnecessary concern)
  • • False Negative: Low prediction, layoff occurs (cost: missed warning)

We optimize for recall (catching actual layoffs) while maintaining high precision (avoiding false alarms).

Continuous Improvement

The model is retrained weekly with new data. Performance metrics are tracked in real-time, and the prediction threshold is adjusted to maintain ≥80% accuracy.