Dimensionality Reduction Evolution and Validation

Abstract

In this paper, proposing visualized and quantitative evaluation methods for validation dimensionality reduction techniques performance. Four well known techniques for dimensionality reduction evaluated, verify the capacity of generating a lower dimensional chemical space with minimum information error. Real chemical database used to generate a sample with specific structure as an input to evolution process. The evaluation is performed in two ways: first visually checking the sample structure, second measuring the local and global distance based error rate that are trained on the low-dimensional data representation. All the evaluation have been done for the four dimensional reduction techniques then result shows weather "trustable" to technique that likely preserves properties with low generalization error and "Un-trustable" which does not preserve the properties.