An Outlier Detection Algorithm Based on Spectral Clustering

Outlier detection is widely used for many areas such as credit card fraud detection, discovery of criminal activities in electronic commerce, weather prediction and marketing. In this paper, we demonstrate the effectiveness of spectral clustering in dataset with outliers. Through spectral method we can use the information of feature space with eigenvectors rather than that of the whole dataset to obtain stable clusters. Then we introduce the cluster-based local outlier factor to identify and find the outliers in dataset. The experimental results show that our outlier detection algorithm outperforms the K-means based algorithm with high precision and low false alarm rate as well as desirable coverage ratio.