Numerous real-world applications are hampered by the fact that typical anomaly detection methods used in full-dimensional fields perform much worse as dimensionality rises. The method for choosing an important feature subspace and performing anomaly detection in the associated subspace projection is proposed in this study. Maintaining detection accuracy under high-dimensional conditions is the goal. The proposed method determines the angle between each pair of two lines for a particular anomaly candidate: the first line is joined by the relevant data point and the centers of its adjacent points, and the second line is one of the axis-parallel lines. The candidate's axis-parallel subspace is thus made up of those dimensions that have a comparatively modest angle with the initial line. An further experiment using an industrial dataset showed how the suggested approach could be used for fault detection jobs and emphasized one of its benefits, which is the capacity to provide a first interpretation of abnormality through feature ordering in pertinent subspaces.

Big Data Anomalies, Fault Detection, Agile Methodologies