machine learning model / in development
Claims Recovery Model
A model that estimates which freight claims are worth pursuing, so recovery effort goes where it is most likely to pay off.
the problem.
Not every claim is worth the time it takes to file. Recovery teams need to know which ones to chase first.
what I built.
A predictive model trained on historical claims. The hard part was not the model, it was making the numbers trustworthy: I worked through data leakage, overfitting, fairness checks, survivorship bias and reproducibility before calling any result real.
Across iterations the classifier landed between 0.84 and 0.93 AUC, with R² between 0.62 and 0.69.
how it works.
- audit the training data for leakage
- build a reproducible training pipeline
- validate against overfitting and survivorship bias
- check fairness across groups
- report results people can trust