A note on accelerating the local outlier factor method on one-dimensional data

Abstract

The local outlier factor (LOF) method, which is proposed by Breunig et al. (2000), is one of the most common techniques to detect outliers or abnormal data points in a dataset. It compares the density of a data point with the densities of its k-nearest neighbors. This study presents an algorithm to perform LOF much faster than conventional methods, especially for one-dimensional data. Its worst-case time complex- ity is only O(nk), and space complexity is O(n). The performance is also computation- ally compared with the DMwR package, which implements Breunig et al. (2000) in R language. ICIC International © 2020.