ADINOF: adaptive density summarizing incremental natural outlier detection in data stream

Outlier detection in the stream data has emerged as the challenging problems with the precipitously growing demand of applications like intrusion detection, sensor malfunctioning, fraud detection, and system failures, etc. To address these problems many density-based algorithms have been proposed for detecting the outliers in stream data. Still, it suffers a serious problem with the degree of outlierness measures on its neighbors. Using a right number k is not straightforward in the stream data since we do not know prior distribution of neighbor's points. And, it must improvise with incoming data as appeared in the stream data. Additionally, stream-based algorithms are not able to detect sequential outliers as well as are having memory constraints. These challenges motivate the authors to propose Self-Adaptive Density Summarizing incremental Natural Outlier Detection in Data Stream (ADINOF) with skipping scheme and without skipping scheme (ADINOF_NS) that successfully overcome the challenges. Our comprehensive experimental evaluations demonstrate that ADINOF and ADINOF_NS significantly outperforms the competitive executed algorithms (iLOF, MiLOF, TADILOF, DILOF, and DILOF_NS).

Automated summary

This paper discusses the issue of outlier detection in stream data and proposes a new Self-Adaptive Density Summarizing incremental Natural Outlier Detection in Data Stream (ADINOF) algorithm, which successfully addresses some of the challenges faced by traditional algorithms.