Scalable fuzzy multivariate outliers identification towards big data applications

Abstract

Data outliers is intrinsically a fuzzy concept and should be treated as such. This paper is a continuation of a research on fuzzy outliers. Extending the BACON algorithm, FBACON1 and FBACON2 have been proposed as fuzzy solutions to the crisp decision boundary of BACON. This paper investigates the scalability potentials and drawbacks of FBACON1 and FBACON2 in Big Data. The investigation concluded that the sensitivity of FBACON2 towards Big Data. Therefore, this paper introduces FBACON3 as a more scalable solution than FBACON2. Three performance measures have been introduced to compare the performance of the three solutions. The study shows that FBACON1 provided the best computational performance followed by FBACON3. However, in terms of the other two measures FBACON2 and FBACON3 are tied but they outperformed FBACON1. Considering the sensitivity towards Big Data volumes and the computation time, FBACON3 is a better candidate than FBACON2. Code metadataPermanent link to reproducible Capsule: <https://doi.org/10.24433/CO.6812430.v1; >.