A Comparative Study of Local Outlier Factor Algorithms for Outliers Detection in Data Streams

Abstract

Outlier detection analyzes data, finds out anomalies, and helps to discover unforeseen activities in safety crucial systems. Outlier detection helps in early prediction of various fraudulent activities like credit card theft, fake insurance claim, tax stealing, real-time monitoring, medical systems, online transactions, and many more. Detection of outliers in the data streams is extremely challenging as compared to static data since data streams are continuous, highly changing and unending in nature. Density-based outlier detection using local outlier factor (LOF) is the prominent method for detecting the outliers in the data streams. In this paper, we provide an insight on outlier detection and various challenges involved while detection of outliers in the data streams. We concentrate on density-based outlier detection and review major local outlier factor (LOF) based outlier detection algorithms in details. We also perform a comparative study of existing LOF algorithms to evaluate the performance in term of several parameters.