Adaptive Feature-Space Conformal Transformation for Imbalanced-Data Learning

When the training instances of the target class are heavily outnumbered by non-target training instances, SVMs can be ineffective in determining the class boundary. To remedy this problem, we propose an adaptive conformal transformation (ACT) algorithm. ACT considers feature-space distance and the class-imbalance ratio when it performs conformal transformation on a kernel function. Experimental results on UCI and real-world datasets show ACT to be effective in improving class prediction accuracy.

• Type: Conference
• Year: 2003
• Where: ICML
• Authors: Gang Wu, Edward Y. Chang
• Keywords: Adaptive Conformal Transformation, Class Prediction Accuracy, ICML 2003, Machine Learning, Non-target Training Instances