Bandwidth selection for kernel-based classification
Ofir Lindenbaum, Arie Yeredor, Amir Z. Averbuch · 2016
Dimensionality reduction is an essential step in various machine learning tasks. Applying classification algorithms to the reduced space is often more efficient and accurate. We focus on kernel based dimensionality reduction techniques, and propose to set the bandwidth such that a coherent mapping is extracted. The proposed framework is simulated on artificial and real dataset, results show a high correlation between optimal classification rates and the proposed bandwidth.