Visualization and Anomaly Detection in Quantitative Data: A 3D Approach Using Metric Encoding

Abstract

The article proposes a new hybrid model for detecting anomalies in continuous quantitative data represented through 3D space, a very important function as far as micro services and cloud computing are concerned. Such 2D dashboards typically need an overt reliance on huge visual codes like color and size resulting in visual cluttering, color blindness, and high cognitive load. The solution integrated deep metric encoding and TDA. Specifically, the metric encoding part leveraged deep neural networks to convert data into a reduced dimensional space where normal data were grouped together and anomalies could be separated based on distance; while the topological structure of encoded data was analyzed for anomalies through persistent homology which might disrupt global or local geometry of data such as isolated loops that cannot be detected by classical distance-based techniques. The hybrid algorithm combines these two criteria using a weighted sum to calculate a final anomaly score.