This has files from the Allstate Severity Prediction Kaggle competition.

The objective was to predict the severity of claims with the error being measured by the mean absolute error. Because of the focus on the mean absolute error instead of mean root sum of squares error, there is less emphasis on extreme outliers (since their squared error is becomes much more extreme in mean root sum of squares).

An overview of my methods are described in the powerpoint deck in the main directory.