Jason M. Klusowski, PhD
Assistant Professor of Operations Research & Financial Engineering, Princeton University
Classification imbalance, where one class is much rarer than the other, is a pervasive challenge in data analysis. This talk views classification imbalance as a transfer learning problem, with models trained on imbalanced source data but evaluated under a balance target distribution. Within this framework, common oversampling methods such as bootstrapping and SMOTE are compared by examining how well they account for differences between training and evaluation distributions. The results show that simple bootstrapping can outperform SMOTE in moderate to high dimensions, offering practical guidance for choosing oversampling strategies in classification imbalance problems.
The Department of Epidemiology & Biostatistics welcomes all participants to our events. If you need a reasonable accommodation to participate in this event because of a disability, please contact Liz Buggs ([email protected]).