Class imbalance remains a common obstacle in tabular machine learning, where one class dominates the training data. A new preprint on arXiv examines weighting-based strategies for addressing this problem, specifically in the context of classification and regression trees (CART). The authors compare the naive, unweighted baseline against rebalancing approaches that assign different weights to classes.

According to the abstract, the work centers on the dynamics of these weighting strategies and their local peak behavior in CART trees. However, the released abstract cuts off mid-sentence, so the full findings and conclusions are not yet available from this source. Readers interested in the details will need to consult the full paper.