LSTM-Based Anomaly Detection: Detection Rules from Extreme Value Theory

Abstract

In this paper, we explore various statistical techniques for anomaly detection in conjunction with the popular Long Short-Term Memory (LSTM) deep learning model for transportation networks. We obtain the prediction errors from an LSTM model, and then apply three statistical models based on (i) the Gaussian distribution, (ii) Extreme Value Theory (EVT), and (iii) the Tukey's method. Using statistical tests and numerical studies, we find strong evidence against the widely employed Gaussian distribution based detection rule on the prediction errors. Next, motivated by fundamental results from Extreme Value Theory, we propose a detection technique that does not assume any parent distribution on the prediction errors. Through numerical experiments conducted on several real-world traffic data sets, we show that the EVTbased detection rule is superior to other detection rules, and is supported by statistical evidence.

A Framework for End-to-End Deep Learning-Based Anomaly Detection in Transportation Networks

“…Further, in our recent work [24], after comparing different detection strategies for hybrid deep anomaly detection, we noticed the potential of a strategy based on extreme values. We found that an EVT-based detection rule performed better than other popular detection techniques.…”

“…We conclude our work in Section VII. 1 A part of this work has been presented as a conference paper [24].…”

Improved Data-Driven Root Cause Analysis in a Fog Computing Environment

“…The size of sliding window n, n < M, where M is size of data. The m*k inputs are passed to LSTM network (Vega García & Aznarte, 2020, Davis et al, 2019) to predict the next value, say x(1). Initially, window contains {x1, x2,…, xm}inputs are feed to LSTM network and it predicts xm+1, at the second step {x2, x3,…, xm+1} are fed to network to get xm+2 as the prediction value.…”

“…The various existing works are compared to the proposed model to evaluate the performance of the model. Machine learning models for multi-step time series prediction includes LSTM model [21], LSTM autoencoder model (Davis et al, 2019). Further, to compare the proposed model with the existing model, the parameters computational power, resource consumption, the delay, CPU, and energy consumption are used.…”