Autoencoder with Gramian Angular Summation Field for Anomaly Detection in Multivariate Time Series Data

Abstract

Uncertainty is ubiquitous in data and constitutes a challenge in real-life data analysis applications. To deal with this challenge, we propose a novel method for detecting anomalies in time series data based on the Autoencoder method, which encodes a multivariate time series as images by means of the Gramian Angular Summation Field (GASF). Multivariate time series data is represented as 2D image data to enhance the performance of anomaly detection. The proposed method is validated with four time-series data sets. Experimental results show that our proposed method can improve validity and accuracy on all criteria. Therefore, effective anomaly detection in multivariate time series data can be achieved by combining the methods of Autoencoder and Gramian Angular Summation Field.