Calibrated Unsupervised Anomaly Detection in Multivariate Time-series using Reinforcement Learning

TL;DR

This paper presents a novel unsupervised anomaly detection method for multivariate time-series data that combines reinforcement learning, wavelet analysis, and a calibrated decision boundary to improve detection accuracy and reduce false negatives.

Contribution

It introduces a reinforcement learning framework in the latent space of an autoencoder, enhanced with wavelet analysis and a supervised calibration step for better anomaly detection.

Findings

Improved detection of both sudden and subtle anomalies.

Reduced false negatives through calibrated decision boundaries.

Enhanced multi-resolution anomaly detection using wavelet features.

Abstract

This paper investigates unsupervised anomaly detection in multivariate time-series data using reinforcement learning (RL) in the latent space of an autoencoder. A significant challenge is the limited availability of anomalous data, often leading to misclassifying anomalies as normal events, thus raising false negatives. RL can help overcome this limitation by promoting exploration and balancing exploitation during training, effectively preventing overfitting. Wavelet analysis is also utilized to enhance anomaly detection, enabling time-series data decomposition into both time and frequency domains. This approach captures anomalies at multiple resolutions, with wavelet coefficients extracted to detect both sudden and subtle shifts in the data, thereby refining the anomaly detection process. We calibrate the decision boundary by generating synthetic anomalies and embedding a supervised framework within the model. This supervised element aids the unsupervised learning process by fine-tuning the decision boundary and increasing the model's capacity to distinguish between normal and anomalous patterns effectively.