SLICE: A Novel Method to Find Local Linear Correlations by Constructing Hyperplanes

Finding linear correlations in dataset is an important data mining task, which can be widely applied in the real world. Existing correlation clustering methods combine clustering with PCA to find correlation clusters in dataset. These methods may miss some correlations when instances are sparsely distributed. Previous studies are limited to find the primary linear correlation of the dataset. However, there may be some interesting local linear correlations exist in data subsets. This paper develops an efficient approach to seek multiple local linear correlations in dataset. The main contributions of this paper are: (1) analyzing the limitations of applying current methods on finding linear correlations in data subsets; (2) developing a novel algorithm, SLICE (significant local linear correlation searching), to find multiple local linear correlations in data subsets. The basic idea of SLICE is using a heuristic way to construct hyperplanes that represent linear correlations; (3) conducti...

• Type: Conference
• Year: 2009
• Where: WAIM
• Authors: Liang Tang, Changjie Tang, Lei Duan, Yexi Jiang, Jie Zuo, Jun Zhu
• Keywords: Information Management, Linear Correlations, Local Linear Correlations, Multiple Local Linear, WAIM 2009