LSTM-GAN for Enhanced Anomaly Detection in Time Series Data

Abstract

Anomaly detection (AD) is crucial for identifying abnormal patterns in time series data that deviate from expected behavior, potentially preventing serious failures. Traditional methods, such as Statistical Process Control (SPC), require extensive human knowledge for setting model assumptions. Unsupervised machine learning approaches have since evolved, employing clustering and predictive modeling to identify anomalies. This paper introduces the LSTM-GAN model, which leverages Long Short-Term Memory (LSTM) networks as the backbone of both the generator and discriminator, and utilizes Wasserstein distance for improved measurement accuracy. Tested on several real-world datasets, the proposed model demonstrates superior performance in detecting anomalies in time series data.