An Information Distance Metric Preserving Projection Algorithm
This paper proposes a novel dimensionality reduction algorithm. The algorithm, coined information distance metric preserving projection (IDPP), aims to identify the complicated intrinsic property of high dimensional space. IDPP employed geodesic information distance to evaluate the relationship between each pair-wise data points. It yielded a distance preserving projection to map sample data from high dimensional observation space to low dimensional feature one. IDPP preserves intrinsic structure of high dimensional space globally. It possesses explicit projection formula which makes it easily to be used for new sample data. Unsupervised and supervised approaches constructed on the basis of IDPP was evaluated on financial data. Experimental results show that trustworthiness of IDPP is almost the same as ISOMAP, and it performs much better than the rival algorithms.
KeywordsDimensionality reduction Manifold learning Financial analysis
This work is supported partly by the Beijing Social Science Foundation (NO. 16SRB021), Beijing Philosophy and Social Science Foundation (NO. 16YJB029).
- 1.Moyer, R.C., McGuigan, J., Rao, R.: Contemporary Financial Management. Cengage Learning, Ohio (2012)Google Scholar
- 11.de Silva, V., Tenenbaum, J.B.: Global versus local methods in nonlinear dimensionality reduction. In: Advances in Neural Information Processing Systems, vol. 15, pp. 705–712 (2003)Google Scholar
- 12.Cai, D., He, X., Han, J.: Spectral regression for efficient regularized subspace learning. In: ICCV 2007 (2007)Google Scholar
- 13.Venna, J., Kaski, S.: Local multi-dimensional scaling with controlled trade-off between trustworthiness and continuity. In: Workshop of Self-Organizing Maps, pp. 695–702 (2005)Google Scholar