EcoSta 2019: Start Registration
View Submission - EcoSta2019
A0576
Title: Cluster-based multiclass linear discriminant analysis Authors:  Kei Hirose - Kyushu University (Japan) [presenting]
Kanta Miura - Kyushu University (Japan)
Atori Koie - Nissan Motor Company Limited (Japan)
Abstract: Multiclass linear discriminant analysis (LDA) is a well-known supervised learning based on a dimensionality reduction technique. In practice, there exists datasets which consist of a large number of classes. In many cases, some of the classes are easy to be classified, and some of them are difficult. In such a case, the LDA can lead to a large classification error, because the data in two similar classes are not appropriately projected onto a low-dimensional space. To handle this issue, we introduce a cluster-based LDA, in which the data in similar classes are categorized as one cluster and then the LDA are conducted to each cluster.