PROJECT TITLE :

A Novel Outlier Detection Method for Multivariate Data

ABSTRACT:

The process of identifying anomalous objects within a set of data has a wide variety of applications in the real world. The vast majority of algorithms for detecting outliers have hidden assumptions and limitations, despite the fact that there is a large number of these algorithms. A novel and effective outlier learning algorithm is proposed in this paper. It is based on the decomposition of the full attributes space into various combinations of subspaces. In this algorithm, the 3D-vectors that represent the data points in each 3D-subspace are rotated about the geometric median using Rodrigues' rotation formula in order to construct the overall outlying score. The method that has been suggested does not involve any parameters, does not call for any distribution assumptions, and is simple to put into practice. Extensive experimental research and comparisons are carried out using six well-known outlier detection algorithms, each originating from a distinct category, on datasets that are both simulated and taken from the real world. The comparison is assessed using the precision @s metric, the average precision metric, the rank power metric, the area under the curve (AUC) metric, and the time complexity metric. The findings indicate that the performance of the method that was proposed is comparable to other methods and shows promise.