![IconGems-Print](images/IconGems-Print.png)
253 – Contributed Poster Presentations: Section on Statistical Computing
Stratified Over Representative K-Folds Cross-Validation
William Franz Lamberti
George Mason University
A method for handling small sample cases using a variation of stratified k-folds crossvalidation is presented. The key difference between traditional stratified k-folds crossvalidation and the sampling approach presented here is over representing the smaller strata. Further, the specific cases to utilize the new approach is when stratified k-folds crossvalidation cannot be used. An intuitive explanation is provided alongside simulations using synthetic data and the famous Fisher or Anderson iris dataset.