A0667
Title: Hierarchical multiclass discriminant analysis via cross-validation
Authors: Kei Hirose - Kyushu University (Japan) [presenting]
Kanta Miura - Kyushu University (Japan)
Abstract: A novel cluster-based LDA is proposed that significantly improves both prediction accuracy and interpretability. We employ hierarchical clustering, and the dissimilarity measure of the two clusters is defined by the cross-validation (CV) value. Therefore, the clusters are constructed such that the error rate is minimized. Our proposed approach requires heavy computational loads because the CV value must be computed at each step of the hierarchical clustering algorithm. We construct an efficient algorithm that computes a consistent estimator of the CV to address the computational issue. The performance of our proposed method is investigated through the application to both artificial and real datasets.
