Canonical Correlation Analysis on IoT Sensor Data
Loading...
Authors
Rao, Vignesh
Date
Type
thesis
Language
eng
Keyword
Canonical Correlation Analysis , Correlation Analysis , Anomaly detection , Convolutional Neural Networks , Autonomous Trains
Alternative Title
Abstract
The rapid increase in the use of sensors to collect continuous data from systems about their operations demands for new analytics systems that can predict anomalies in the data. The numerous connected devices referred to as the Internet of Things (IoT), offer a new challenge to estimate relations among multiple data items, which can help in understanding operational behaviour in IoT systems. While the first ever proposed method for performing CCA (Hotelling, 1992) used eigen-vector calculations and could detect linear correlations, researchers have since proposed various methods to perform CCA using probabilistic, sparse, kernel, and discriminative techniques. Recently, deep learning based CCA methods have been proposed that can determine highly complex nonlinear transformations in data. In this research, we explore two methods - a linear CCA algorithm and a non-linear deep learning based CCA method, and present insights about what the methods concur regarding the contribution of the data features to the overall correlation among two sets of variables. The experiments demonstrate the value of CCA by applying and evaluating the methods on a real life autonomous train dataset collected from a real-time train signalling system equipped with many sensors in New York City over a three-month period. With the evidence of the strength of deep learning based CCA on time series data, we propose a Convolutional Neural Network (CNN) based CCA method. We use the CNN for generating correlation based mappings and statistical anomaly detection methods to develop a collective anomaly detection approach. The proposed method has been validated with a synthetic dataset that is commonly used in research on CCA. We also use an autonomous train positioning system simulation dataset from our industry collaborator. The results indicate strong collective anomaly detection performance with an F1-score of 89.0% and a sensitivity of 94.1%, which can pave the way for application of the proposed models for real time collective anomaly detection and CCA in IoT systems.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.