Introduction to statistical pattern recognition (2nd ed.)
Introduction to statistical pattern recognition (2nd ed.)
Laplacian Eigenmaps for dimensionality reduction and data representation
Neural Computation
Information-theoretic metric learning
Proceedings of the 24th international conference on Machine learning
Feature extraction based on Laplacian bidirectional maximum margin criterion
Pattern Recognition
Distance Metric Learning for Large Margin Nearest Neighbor Classification
The Journal of Machine Learning Research
Efficient and robust feature extraction by maximum margin criterion
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
This paper presents Enhanced Eigenspace Separation Transform (EEST), a novel supervised dimensionality reduction technique for classification. EEST is motivated from the Eigenspace Separation Transform (EST). The criterion of EEST is to maximize the difference in the average lengths of vectors in the underlying two classes, and minimize the intra-class average distances, to improve the generalization capacity of a classifier. We propose upper bounds of the criterion, and a specific solution space to attain these bounds. Existence of such a solution is restricted, thereby we have considered the orthonormal space of the upper bounds in order to achieve better dimensionality reduction, and improve the generalization accuracy of the classifier. A simple Nearest Neighborhood (NN) classification approach is adopted for classification to highlight the novelty of the proposed scheme. Theoretical analysis of the proposed techniques is also carried out. Different synthetic data sets have been used to evaluate the advantages of EEST over EST. Extensive empirical studies are made; and the proposed method is compared with three closely related schemes using real-world data sets to verify the efficiency of the proposed method.