Explainable Outlier Detection for Interval-valued Data

Explainability is increasingly recognized as a key aspect of outlier detection.

However, for complex data structures such as interval-valued data, it remains largely unexplored.

Building on an outlier detection framework based on the Interval Minimum Covariance Determinant estimator, we propose a novel approach to explain the outlyingness of interval-valued observations using the concept of the Shapley value.

We derive a closed-form expression for the Shapley value of the squared robust Interval-Mahalanobis distance, enabling efficient computation of variable contributions.

This formulation allows for a fine-grained interpretation of outliers, providing a detailed decomposition into contributions from centers, ranges, and cross-terms of the interval-valued observations.

Moreover, the Shapley value is closely connected to the concept of cellwise outliers, as it can help identify variable-specific outliers that may not be evident at multivariate level.

We further extend the framework through the Shapley interaction index to capture pairwise variable interactions driving atypical behavior.

The practical utility of the proposed approach is illustrated through two real-world datasets.