Explanation of Gas Turbine Generator Anomaly Detection Using Contextual Outlier Interpretation with Ordering Points to Identify the Clustering Structure

This paper proposes explanation of gas turbine generator anomaly detection using a modified Contextual Outlier Interpretation (COIN) with Ordering Points To Identify the Clustering Structure (OPTICS). For anomaly detection using AI models, users need to be convinced by the explanation for anomaly detection to utilize the AI models. In other words, the application of explainable artificial intelligence (XAI) is important. COIN is one of the appropriate methods for explaining anomaly detection. However, the method requires clustering of normal data in the neighborhood of the target anomaly datum. The original COIN utilizes K-means for clustering, which is time-consuming. Therefore, the authors proposed COIN with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to reduce compution time. However, DBSCAN has a challenge to require manual hyperparameter tuning. OPTICS can generate clusters automatically using changes of Euclidean distance among data and does not require manual hyperparameter tuning. Effectiveness of the proposed COIN with OPTICS based method is verified by comparing with the conventional COIN with K-means and DBSCAN based methods.