|
|
Used an SVM on text to predict fraud from claim notes. This allowed us to automate the work of 15 FTEs.
2014-2016LinkReversed-engineered pre-packaged GLM software, allowing us to automatically produce modeling packets. This reduced a day-long project to minutes.
2014-2016LinkResearched and advised analysts in the company on dimension reduction in auto and credit datasets. We looked into PCA, partial least squares, and lasso regressions.
2014-2016LinkBuilt a customer lifetime value model of our commerical policy data.
2014-2016LinkHelped refresh our decade-old commercial auto model, combining two previous models.
2014-2016LinkI presented research on Shapley values and family errorwise rates that influenced our modeling techniques broadly.
2014-2016Link