Ranking Interesting Subspaces for Clustering High Dimensional Data

摘要

Application domains such as life sciences, e.g. molecular biology produce a tremendous amount of data which can no longer be managed without the help of efficient and effective data mining methods. One of the primary data mining tasks is clustering. However, traditional clustering algorithms often fail to detect meaningful clusters because of the high dimensional, inherently sparse feature space of most real-world data sets. Nevertheless, the data sets often contain clusters hidden in various subspaces of the original feature space. We present a pre-processing step for traditional clustering algorithms, which detects all interesting sub-spaces of high-dimensional data containing clusters. For this purpose, we define a quality criterion for the interestingness of a subspace and propose an efficient algorithm called RIS (Ranking I nteresting Subspaces) to examine all such subspaces. A broad evaluation based on synthetic and real-world data sets empirically shows that RIS is suitable to find all relevant subspaces in large, high dimensional, sparse data and to rank them accordingly.