|
Abstract:
|
A quantile classifier is a classifier based on the distance which corresponds to the quantile. The method is defined by classifying an observation according to the sum of appropriately weighted distances of the components of the observation to the within-class quantiles. This method is effective if the objects are following asymmetric distributions. For example, there is a situation where we classify two classes, one class belongs to the objects following exponential distribution and the other class belongs to the objects following a normal distribution. However, in the case that the observed class labels are corrupted. For example, the label may be flipped with some constant probability. In this situation, the classification accuracy is low. To tackle this problem, we develop the discriminant analysis method using a quantile classifier for corrupted data.
|
