PIXAL: Anomaly Reasoning with Visual Analytics

TL;DR

PIXAL is a visual analytics system designed to assist analysts in reasoning about anomalies by providing pattern detection, trust-building visualizations, and hypothesis validation tools, thereby streamlining the anomaly investigation process.

Contribution

The paper introduces PIXAL, a novel visual analytics system that integrates pattern detection and hypothesis validation tools specifically for anomaly reasoning, developed through iterative collaboration with professional analysts.

Findings

PIXAL helps analysts understand anomalies more effectively.

The system enables generation of actionable hypotheses.

Qualitative study shows improved reasoning efficiency.

Abstract

Anomaly detection remains an open challenge in many application areas. While there are a number of available machine learning algorithms for detecting anomalies, analysts are frequently asked to take additional steps in reasoning about the root cause of the anomalies and form actionable hypotheses that can be communicated to business stakeholders. Without the appropriate tools, this reasoning process is time-consuming, tedious, and potentially error-prone. In this paper we present PIXAL, a visual analytics system developed following an iterative design process with professional analysts responsible for anomaly detection. PIXAL is designed to fill gaps in existing tools commonly used by analysts to reason with and make sense of anomalies. PIXAL consists of three components: (1) an algorithm that finds patterns by aggregating multiple anomalous data points using first-order predicates, (2) a visualization tool that allows the analyst to build trust in the algorithmically-generated predicates by performing comparative and counterfactual analyses, and (3) a visualization tool that helps the analyst generate and validate hypotheses by exploring which features in the data most explain the anomalies. Finally, we present the results of a qualitative observational study with professional analysts. These results of the study indicate that PIXAL facilitates the anomaly reasoning process, allowing analysts to make sense of anomalies and generate hypotheses that are meaningful and actionable to business stakeholders.