Statistical Analysis of using the Shapley Value for Sensor Anomaly Localization with Accurate Classifiers
A new paper analyzes the use of Shapley values for localizing sensor anomalies, comparing their performance against simpler anomaly detection methods. The research proves that for independent sensor observations, the Shapley value approach is equivalent to a less complex method. However, for dependent observations, the Shapley value can be either superior or inferior depending on specific statistical conditions, suggesting potential for combined approaches. AI
IMPACT Provides theoretical insights into anomaly detection methods, potentially improving robustness in sensor systems.