Learning Imbalanced Multi-Class Data with Optimal Dichotomy Weights

Many real data mining tasks suffer from class-imbalance, i.e., some classes have much fewer data than the other classes, whereas the minority classes are more important. Conventional learning methods often try to pursue a high accuracy by assuming that all classes have similar sizes, leading to the fact that the minority class examples are often overlooked and misclassified to the majority classes. However, accuracy is not an adequate evaluation measure when there is class-imbalance because it ignores the larger importance of minority classes.

Provided by: SouthEast SAS Users Group Topic: Big Data Date Added: Oct 2013 Format: PDF

Find By Topic