Anomaly detection and prediction in communication networks using wavelet transforms

Anomaly detection and prediction in communication networks using wavelet transforms. Alarcon-Aquino, V. Ph.D. Thesis, 2003.

Website abstract bibtex 2 downloads

It is important for service providers to monitor their systems in order to detect network anomalies and performance degradations in advance of network/service disruptions. In this regard, several anomaly detection schemes have already been proposed in the literature. These schemes are in most cases based on parametric models and thresholding techniques. The underlying aim of this thesis is to develop models and algorithms based on wavelet transforms for analysing the statistical behaviour of network metrics in order to detect and predict network anomalies and performance degradations in communication networks. The original contributions of this research can be classified into three categories. Firstly, a novel wavelet-based algorithm is proposed for detecting network anomalies in communication networks. The wavelet-based algorithm is then used to detect events in different network metrics of a Dial Internet Protocol service and corporate Proxy servers. A sensor fusion scheme, which combines local decisions made from dispersed wavelet-based sensors, is also investigated. This sensor fusion scheme incorporates the spatial dependencies among the monitored network metrics and hence reduces the number of false alarms generated by each network metric. Secondly, a novel learning algorithm is proposed for time series prediction based on finite impulse response (FIR) neural networks and the multiresolution analysis framework of wavelet theory. A gradient descent method is used to adapt the gain of the non-linear functions in FIR networks at each level of resolution. The multiresolution learning algorithm is compared with previously reported algorithms using a benchmark time series. The algorithm is also applied to network traffic prediction in an Ethernet environment. The results show that the generalisation ability of the FIR network is improved by the multiresolution learning algorithm. Finally, a method is proposed for predicting and monitoring communication network metrics. The proposed method learns to predict the normal behaviour of the monitored network metric and together with an online decision-making algorithm detects and classify deviations from the normal operation region. The proposed method is used to predict and monitor events in corporate Proxy servers and Local/Wide area network traces. Experimental results show that the proposed method is able to identify moderate and severe abnormal network behaviours in advance of reported network faults and thereby providing a useful method for proactively managing communication networks.

@phdthesis{
 title = {Anomaly detection and prediction in communication networks using wavelet transforms},
 type = {phdthesis},
 year = {2003},
 pages = {1-233},
 websites = {https://spiral.imperial.ac.uk/handle/10044/1/11475},
 publisher = {Imperial College London, UK},
 city = {London},
 institution = {Imperial College London},
 department = {Electrical and Electronic Engineering},
 id = {92c40961-3160-36e6-8fb2-c4ce1cd111a1},
 created = {2022-08-29T17:43:47.402Z},
 accessed = {2021-10-23},
 file_attached = {false},
 profile_id = {940dd160-7d67-3a5f-b9f8-935da0571367},
 group_id = {92fccab2-8d44-33bc-b301-7b94bb18523c},
 last_modified = {2024-05-28T18:20:28.883Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 source_type = {THES},
 user_context = {PhD Thesis},
 private_publication = {false},
 abstract = {It is important for service providers to monitor their systems in order to detect network anomalies and performance degradations in advance of network/service disruptions. In this regard, several anomaly detection schemes have already been proposed in the literature. These schemes are in most cases based on parametric models and thresholding techniques. The underlying aim of this thesis is to develop models and algorithms based on wavelet transforms for analysing the statistical behaviour of network metrics in order to detect and predict network anomalies and performance degradations in communication networks. The original contributions of this research can be classified into three categories. Firstly, a novel wavelet-based algorithm is proposed for detecting network anomalies in communication networks. The wavelet-based algorithm is then used to detect events in different network metrics of a Dial Internet Protocol service and corporate Proxy servers. A sensor fusion scheme, which combines local decisions made from dispersed wavelet-based sensors, is also investigated. This sensor fusion scheme incorporates the spatial dependencies among the monitored network metrics and hence reduces the number of false alarms generated by each network metric. Secondly, a novel learning algorithm is proposed for time series prediction based on finite impulse response (FIR) neural networks and the multiresolution analysis framework of wavelet theory. A gradient descent method is used to adapt the gain of the non-linear functions in FIR networks at each level of resolution. The multiresolution learning algorithm is compared with previously reported algorithms using a benchmark time series. The algorithm is also applied to network traffic prediction in an Ethernet environment. The results show that the generalisation ability of the FIR network is improved by the multiresolution learning algorithm. Finally, a method is proposed for predicting and monitoring communication network metrics. The proposed method learns to predict the normal behaviour of the monitored network metric and together with an online decision-making algorithm detects and classify deviations from the normal operation region. The proposed method is used to predict and monitor events in corporate Proxy servers and Local/Wide area network traces. Experimental results show that the proposed method is able to identify moderate and severe abnormal network behaviours in advance of reported network faults and thereby providing a useful method for proactively managing communication networks.},
 bibtype = {phdthesis},
 author = {Alarcon-Aquino, Vicente}
}

Downloads: 2

{"_id":"FpgKWPJeMpb4T7fsS","bibbaseid":"alarconaquino-anomalydetectionandpredictionincommunicationnetworksusingwavelettransforms-2003","author_short":["Alarcon-Aquino, V."],"bibdata":{"title":"Anomaly detection and prediction in communication networks using wavelet transforms","type":"phdthesis","year":"2003","pages":"1-233","websites":"https://spiral.imperial.ac.uk/handle/10044/1/11475","publisher":"Imperial College London, UK","city":"London","institution":"Imperial College London","department":"Electrical and Electronic Engineering","id":"92c40961-3160-36e6-8fb2-c4ce1cd111a1","created":"2022-08-29T17:43:47.402Z","accessed":"2021-10-23","file_attached":false,"profile_id":"940dd160-7d67-3a5f-b9f8-935da0571367","group_id":"92fccab2-8d44-33bc-b301-7b94bb18523c","last_modified":"2024-05-28T18:20:28.883Z","read":false,"starred":false,"authored":false,"confirmed":"true","hidden":false,"source_type":"THES","user_context":"PhD Thesis","private_publication":false,"abstract":"It is important for service providers to monitor their systems in order to detect network anomalies and performance degradations in advance of network/service disruptions. In this regard, several anomaly detection schemes have already been proposed in the literature. These schemes are in most cases based on parametric models and thresholding techniques. The underlying aim of this thesis is to develop models and algorithms based on wavelet transforms for analysing the statistical behaviour of network metrics in order to detect and predict network anomalies and performance degradations in communication networks. The original contributions of this research can be classified into three categories. Firstly, a novel wavelet-based algorithm is proposed for detecting network anomalies in communication networks. The wavelet-based algorithm is then used to detect events in different network metrics of a Dial Internet Protocol service and corporate Proxy servers. A sensor fusion scheme, which combines local decisions made from dispersed wavelet-based sensors, is also investigated. This sensor fusion scheme incorporates the spatial dependencies among the monitored network metrics and hence reduces the number of false alarms generated by each network metric. Secondly, a novel learning algorithm is proposed for time series prediction based on finite impulse response (FIR) neural networks and the multiresolution analysis framework of wavelet theory. A gradient descent method is used to adapt the gain of the non-linear functions in FIR networks at each level of resolution. The multiresolution learning algorithm is compared with previously reported algorithms using a benchmark time series. The algorithm is also applied to network traffic prediction in an Ethernet environment. The results show that the generalisation ability of the FIR network is improved by the multiresolution learning algorithm. Finally, a method is proposed for predicting and monitoring communication network metrics. The proposed method learns to predict the normal behaviour of the monitored network metric and together with an online decision-making algorithm detects and classify deviations from the normal operation region. The proposed method is used to predict and monitor events in corporate Proxy servers and Local/Wide area network traces. Experimental results show that the proposed method is able to identify moderate and severe abnormal network behaviours in advance of reported network faults and thereby providing a useful method for proactively managing communication networks.","bibtype":"phdthesis","author":"Alarcon-Aquino, Vicente","bibtex":"@phdthesis{\n title = {Anomaly detection and prediction in communication networks using wavelet transforms},\n type = {phdthesis},\n year = {2003},\n pages = {1-233},\n websites = {https://spiral.imperial.ac.uk/handle/10044/1/11475},\n publisher = {Imperial College London, UK},\n city = {London},\n institution = {Imperial College London},\n department = {Electrical and Electronic Engineering},\n id = {92c40961-3160-36e6-8fb2-c4ce1cd111a1},\n created = {2022-08-29T17:43:47.402Z},\n accessed = {2021-10-23},\n file_attached = {false},\n profile_id = {940dd160-7d67-3a5f-b9f8-935da0571367},\n group_id = {92fccab2-8d44-33bc-b301-7b94bb18523c},\n last_modified = {2024-05-28T18:20:28.883Z},\n read = {false},\n starred = {false},\n authored = {false},\n confirmed = {true},\n hidden = {false},\n source_type = {THES},\n user_context = {PhD Thesis},\n private_publication = {false},\n abstract = {It is important for service providers to monitor their systems in order to detect network anomalies and performance degradations in advance of network/service disruptions. In this regard, several anomaly detection schemes have already been proposed in the literature. These schemes are in most cases based on parametric models and thresholding techniques. The underlying aim of this thesis is to develop models and algorithms based on wavelet transforms for analysing the statistical behaviour of network metrics in order to detect and predict network anomalies and performance degradations in communication networks. The original contributions of this research can be classified into three categories. Firstly, a novel wavelet-based algorithm is proposed for detecting network anomalies in communication networks. The wavelet-based algorithm is then used to detect events in different network metrics of a Dial Internet Protocol service and corporate Proxy servers. A sensor fusion scheme, which combines local decisions made from dispersed wavelet-based sensors, is also investigated. This sensor fusion scheme incorporates the spatial dependencies among the monitored network metrics and hence reduces the number of false alarms generated by each network metric. Secondly, a novel learning algorithm is proposed for time series prediction based on finite impulse response (FIR) neural networks and the multiresolution analysis framework of wavelet theory. A gradient descent method is used to adapt the gain of the non-linear functions in FIR networks at each level of resolution. The multiresolution learning algorithm is compared with previously reported algorithms using a benchmark time series. The algorithm is also applied to network traffic prediction in an Ethernet environment. The results show that the generalisation ability of the FIR network is improved by the multiresolution learning algorithm. Finally, a method is proposed for predicting and monitoring communication network metrics. The proposed method learns to predict the normal behaviour of the monitored network metric and together with an online decision-making algorithm detects and classify deviations from the normal operation region. The proposed method is used to predict and monitor events in corporate Proxy servers and Local/Wide area network traces. Experimental results show that the proposed method is able to identify moderate and severe abnormal network behaviours in advance of reported network faults and thereby providing a useful method for proactively managing communication networks.},\n bibtype = {phdthesis},\n author = {Alarcon-Aquino, Vicente}\n}","author_short":["Alarcon-Aquino, V."],"urls":{"Website":"https://spiral.imperial.ac.uk/handle/10044/1/11475"},"biburl":"https://bibbase.org/service/mendeley/940dd160-7d67-3a5f-b9f8-935da0571367","bibbaseid":"alarconaquino-anomalydetectionandpredictionincommunicationnetworksusingwavelettransforms-2003","role":"author","metadata":{"authorlinks":{}},"downloads":2},"bibtype":"phdthesis","biburl":"https://bibbase.org/service/mendeley/940dd160-7d67-3a5f-b9f8-935da0571367","dataSources":["uTykyhFv6T7J2dX6z","nyWSrx2yzJEsoe49A"],"keywords":[],"search_terms":["anomaly","detection","prediction","communication","networks","using","wavelet","transforms","alarcon-aquino"],"title":"Anomaly detection and prediction in communication networks using wavelet transforms","year":2003,"downloads":2}