Please use this identifier to cite or link to this item: https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4189
Title: LSTM based Framework for Time Series Anomaly Detection
Authors: Eranga, N.A.A.H.
Keywords: Anomaly detection
Long short-term memory networks
Convolutional neural networks
Dynamic time warping
hybrid neural networks
Issue Date: 22-Jul-2021
Abstract: Anomaly detection also known as Outlier detection is identification of data points, items or events that does not fit the expected behavior. In time series anomaly detection the objective is to detect anomalies in temporal data, a series of data points indexed in time order. This data can be network data, spatio temporal data and stream data, etc. In this dissertation, an approache is proposed to detect anomalies in time series data based on deep neural networks and dynamic time warping algorithm. Initially we use a long short-term memory network to predict time series data. Then combined the LSTM network predictions with a convolutional neural network to get more accurate predictions. The predictions are compared with original time series data in a time window approach using DTW algorithm to identify the anomalies within the time series. Evaluation process of the model is done using several steps. In the initial steps when developing the model three datasets are used to evaluate and test the models. For the final evaluation numenta anomaly benchmark dataset a novel benchmark for evaluating algorithms for anomaly detection, is used.
URI: http://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4189
Appears in Collections:2018

Files in This Item:
File Description SizeFormat 
2014CS033.pdf3.76 MBAdobe PDFView/Open


Items in UCSC Digital Library are protected by copyright, with all rights reserved, unless otherwise indicated.