Consistent Robust Analytical Approach for Outlier Detection in Multivariate Data using Isolation Forest and Local Outlier Factor

Abstract

Outlier detection in real-time from multivariate streaming data is an important research subject in numerous areas. The new presentation of gradual Neighborhood Anomaly Variable (iLOF) and its variations has acquired consideration for their high recognition execution in information streams with evolving circulations. This paper presents a new intelligent exception location framework called include rich intelligent anomaly discovery, which integrates human interaction into the detection process to improve performance and simplify detection. It offers interactive mechanism that allows for instinctive client cooperation during every vital stage of the fundamental anomaly recognition calculation, for example, thick cell choice, area mindful distance thresholding, and last top exception approval. This approach helps resolve the challenge of specifying key parameters like density and distance thresholds in other outlier detection methods. Additionally, the system proposes an innovative optimization method to enhance grid-based space partitioning. The Local Outlier Factor (LOF) measures the level of outlierness of each occasion in light of the conveyance thickness in the dataset. Higher LOF values indicate a higher likelihood of an instance being an outlier. Instances with LOF values above a set threshold are identified as outliers. The calculation of LOF values involves several steps, which are detailed in original articles for further reference.