Accelerating LOF Outlier Detection Approach

Abstract

Outliers are the deformities in the data that diverges from the normal behavior. Detection of outlier points is a crucial task as it leads to the extraction of the discordant observations in different domains. One of the most popular density-based outlier detection techniques is local outlier factor (LOF), and later many variants of this approach are also introduced. These techniques have more execution time as they calculate the outlier score for every data point. In this paper, we propose an approach that first detects the data points which have a high probability of being an outlier (i.e., probable outliers) based on Z-score and modified Z-score statistical techniques. Subsequently, we compute the anomaly score of only these probable outliers. Therefore, we avoid to calculate the outlier score of a substantial number of data points. We conducted experiments on synthetic dataset as well as on real-world datasets, and experimental results demonstrate that our proposed approaches outperform the popular outlier detection technique LOF and its variants.