A boosted outlier detection method based on the spectrum of the Laplacian matrix of a graph

TL;DR

This paper introduces a novel outlier detection method leveraging the spectrum of the Laplacian matrix, combining boosting and sparse data techniques to improve efficiency and scalability over existing spectral methods.

Contribution

It presents a new spectrum-based outlier detection algorithm that uses sparse Laplacian matrices and boosting, enabling application to larger datasets.

Findings

Competitive performance on synthetic datasets

Reduced computational burden due to sparsity

Effective for larger datasets compared to spectral clustering

Abstract

This paper explores a new outlier detection algorithm based on the spectrum of the Laplacian matrix of a graph. Taking advantage of boosting together with sparse-data based learners. The sparcity of the Laplacian matrix significantly decreases the computational burden, enabling a spectrum based outlier detection method to be applied to larger datasets compared to spectral clustering. The method is competitive on synthetic datasets with commonly used outlier detection algorithms like Isolation Forest and Local Outlier Factor.