Ordinal Outlier Algorithm for Anomaly Detection of High-Dimensional Data Sets

Concerning with the non-linear characteristics of high dimensional data set, we propose a ordinal outlier algorithm for anomaly detection for high-dimensional data sets, which is combined with the idea of Locally Linear Embedding. The core idea of the method is to perform dimensionality reduction in the local space of high-dimensional data set and further use accumulated value of probabilities to reveal outliers. The algorithm retains local separability of data and also has the advantages of PPCA model, which makes it more suitable for processing high dimensional data. Experimental results demonstrate its high feasibility and efficiency and it has power to find more meaningful outliers.