Study of outlier mining algorithms

A local outlier mining algorithm is put forward based on the partition of subspaces. The algorithm first divides the data set into disjoint subspaces, using the degree of skewness to measure the pros and cons of the space division, and adopting the particle swarm optimization algorithm to search the optimal partition of subspaces set; then aiming at each optimal partition of subspaces to calculate the local outlier factor SPLOF value of its data object, and take the SPLOF value as the local deviation degree of measuring the data object. Finally adopting the discrimination astronomical spectral data as the data set, experiments verify that the algorithm possesses the excellence of not relying on users´ input parameters, strong flexibility, and efficient operation and so on.