情報処理学会 第73回全国大会講演要旨

The Outlier detection plays an important role in data mining or statistics method, which can removed abnormal data points or find some strange behaves in real applications. However, when dimension is very high, these methods seem not effective because of “curse of dimension”. Though some studies such as the subspace clustering dimension optimization are used to solve these problems, complexities in time expense, space occupation or inaccuracy still need to be considered.
In this paper, a novel approach RPGS (Rim Project Grid Statistic) is proposed, with dimension projection and statistic method, each point gets values for each dimension and between different dimensions. The outliers are easily found by comparing their results. The first experiments show that this method is feasible in low dimension and also effective in high dimension.