Code underlying the publication: "PATE: Proximity-Aware Time series anomaly Evaluation"

This repository provides the implementation of Proximity-Aware Time Series Anomaly Evaluation (PATE), a novel metric introduced to address the limitations of existing evaluation methods for time series anomaly detection. PATE incorporates proximity-based weighting with buffer zones around anomaly intervals to account for temporal complexities such as Early or Delayed detections, Onset response time, and Coverage level. It computes a weighted version of the Area Under Precision and Recall curve, offering a more accurate and meaningful assessment of anomaly detection models. Experimental results validate PATE's ability to distinguish performance differences across various models and scenarios.