OIE: An Interpretable System for Outlier Explanation and Summarization

Outlier detection is crucial for preventing financial fraud, network intrusions, and device failures. However, while existing methods excel at identifying outliers, they often fall short of providing clear and interpretable explanations. This limitation forces users to manually analyze numerous outliers, making the process both time-consuming and inefficient. Additionally, current summarization approaches often focus only on data attributes, ignoring that outliers in the same subspace may have different causes. Consequently, these methods produce broad summaries that make quick diagnosis difficult. To address these challenges, we propose OIE, a system that generates interpretable, fine-grained rules to summarize and explain outlier detection results. OIE leverages decision trees to generate concise rules, balancing simplicity and classification accuracy. Additionally, it employs dynamic data partitioning and a boundary stabilizer to efficiently handle high-dimensional and complex datasets. Through multiple real-world scenarios, OIE demonstrates effective anomaly detection and summarization, providing actionable insights and enhancing the efficiency of outlier analysis. A demonstration video of OIE is available at:http://youtu.be/YzfDdF9f5HI.