Advanced Memory Efficient Outlier Detection Approach for Streaming Data using Swarm Optimization

Abstract

Outlier detection techniques detect abnormal behavior in data and are useful in a variety of applications. In a real-life scenario, various applications generate large-scale data every day. Outlier detection over such continuous/streaming data is a challenging task due to its volume and limitations in processing memory. This paper presents an outlier detection approach called Advanced Memory Efficient Outlier Detection (A-MEOD) that is able to find outliers in streaming data in a memory-efficient manner. The outlier detection is based on the MEOD technique and Local Correlation Integral (LOCI) algorithm. Further the A-MEOD technique reduces the LOCI calculations and finds the top M outliers using Knorr’s definition. The results of utilization of A-MEOD are compared with MiLOF and MEOD in terms of accuracy, time, and memory requirements.