Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi
Intrusion detection systems (IDS) are vital to cyber security, particularly with the presence of various networked computer infrastructures. An efficient IDS uses computational methods as techniques of machine learning (ML) to enhance the rates of detection to obtain the lowest false positive rate,...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Online Access: | https://ir.uitm.edu.my/id/eprint/83493/1/83493.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| id |
my-uitm-ir.83493 |
|---|---|
| record_format |
uketd_dc |
| institution |
Universiti Teknologi MARA |
| collection |
UiTM Institutional Repository |
| language |
English |
| advisor |
Hashim, Habibah |
| description |
Intrusion detection systems (IDS) are vital to cyber security, particularly with the presence of various networked computer infrastructures. An efficient IDS uses computational methods as techniques of machine learning (ML) to enhance the rates of detection to obtain the lowest false positive rate, although such rates tend to be reduced by the big amount of irrelevant features as an optimization issue. Data clustering, clustering items from information into significant clusters. Based on the above components and circumstances, many studies have been performed on data clustering problems. Despite attempts to solve the data clustering issues, there are also many variants of modified algorithms in traditional information clustering that attempt to solve issues such as clustering algorithms based on condensation. These algorithms are aimed at overcoming them in terms of offering high accuracy and reduced computational time, the quality of their outcomes still does not fulfill researchers. Moreover, shortage of reliable methods on a new dataset for the intrusion detection system and anomaly detection in terms of classification is an issue. Thus, this study is looking for better, new dataset and more reliable optimized method for detecting the intrusion with highest accuracy. One of the major ML problems is classification and it is believed that many previous researchers did not apply their methods on the latest and updated dataset for testing and validation, therefore the results may no longer be applicable and are not reliable within the current attacks. Further, the shortage of efficient feature selection techniques gives rise to low accuracy in anomaly detection. One of the main steps after the data collection stage of any method is selecting a subset of the features to be used for the feature selection process. Some of the previous researchers used a feature set selection which is introduced for IDS but there still shortage in their detection rate and selected amounts of features. To be able to address the challenges that mentioned above in this study, an architecture is proposed in order to select relevant feature subsets and improve clustering accuracy. In this study, researcher is trying to improve the clustering of data using an efficient technique via Enhanced Binary Particle Swarm Optimization (EBPSO) as feature selection. Also, this research will use the most updated dataset wich called CICIDS2017 that it covers the majority of current intrusion and attacks. This approach that is according to the DNN model reduces irrelevant features in the intrusion detection data sets of CICIDS2017 to improve the accuracy and cluster high-scale data sets. This strategy includes a number of components that are a novel approach to clustering generation. In fact a data clustering method is proposed consisting of separate outputs: (i) To select a relevant subset of original features based on our proposed algorithm; which is Enhanced Binary Particle swarm Optimization (EBPSO), (ii) To mine data using various data chunks (windows) and overcome a failure of single clustering. An experimental analysis is conducted by several experiments to assess the efficiency of the suggested methods that have been tested within the benchmark datasets, namely CICIDS2017. In comparison to different metaheuristic algorithms for feature selection, experimental outcomes indicate that the suggested method is capable of reducing dimensionality cost, the number of irrelevant features and produce reasonable accuracy. Experiments demonstrate and prove that the proposed EBPSO method produces better accuracy mining data and selecting subset of relevant features comparing other algorithms. In addition, experiments prove that the enhanced algorithm shows a higher performance through lower false positive, higher accuracy, and better CPU time. |
| format |
Thesis |
| qualification_name |
Doctor of Philosophy (PhD.) |
| qualification_level |
Doctorate |
| author |
Atefi, Kayvan |
| spellingShingle |
Atefi, Kayvan Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| author_facet |
Atefi, Kayvan |
| author_sort |
Atefi, Kayvan |
| title |
Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| title_short |
Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| title_full |
Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| title_fullStr |
Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| title_full_unstemmed |
Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi |
| title_sort |
enhanced ai-based anomaly detection method in the intrusion detection system (ids) / kayvan atefi |
| granting_institution |
Universiti Teknologi MARA (UiTM) |
| granting_department |
Faculty of Electrical Engineering |
| publishDate |
2019 |
| url |
https://ir.uitm.edu.my/id/eprint/83493/1/83493.pdf |
| _version_ |
1783736370018123776 |
| spelling |
my-uitm-ir.834932023-11-16T08:36:18Z Enhanced AI-based anomaly detection method in the intrusion detection system (IDS) / Kayvan Atefi 2019 Atefi, Kayvan Intrusion detection systems (IDS) are vital to cyber security, particularly with the presence of various networked computer infrastructures. An efficient IDS uses computational methods as techniques of machine learning (ML) to enhance the rates of detection to obtain the lowest false positive rate, although such rates tend to be reduced by the big amount of irrelevant features as an optimization issue. Data clustering, clustering items from information into significant clusters. Based on the above components and circumstances, many studies have been performed on data clustering problems. Despite attempts to solve the data clustering issues, there are also many variants of modified algorithms in traditional information clustering that attempt to solve issues such as clustering algorithms based on condensation. These algorithms are aimed at overcoming them in terms of offering high accuracy and reduced computational time, the quality of their outcomes still does not fulfill researchers. Moreover, shortage of reliable methods on a new dataset for the intrusion detection system and anomaly detection in terms of classification is an issue. Thus, this study is looking for better, new dataset and more reliable optimized method for detecting the intrusion with highest accuracy. One of the major ML problems is classification and it is believed that many previous researchers did not apply their methods on the latest and updated dataset for testing and validation, therefore the results may no longer be applicable and are not reliable within the current attacks. Further, the shortage of efficient feature selection techniques gives rise to low accuracy in anomaly detection. One of the main steps after the data collection stage of any method is selecting a subset of the features to be used for the feature selection process. Some of the previous researchers used a feature set selection which is introduced for IDS but there still shortage in their detection rate and selected amounts of features. To be able to address the challenges that mentioned above in this study, an architecture is proposed in order to select relevant feature subsets and improve clustering accuracy. In this study, researcher is trying to improve the clustering of data using an efficient technique via Enhanced Binary Particle Swarm Optimization (EBPSO) as feature selection. Also, this research will use the most updated dataset wich called CICIDS2017 that it covers the majority of current intrusion and attacks. This approach that is according to the DNN model reduces irrelevant features in the intrusion detection data sets of CICIDS2017 to improve the accuracy and cluster high-scale data sets. This strategy includes a number of components that are a novel approach to clustering generation. In fact a data clustering method is proposed consisting of separate outputs: (i) To select a relevant subset of original features based on our proposed algorithm; which is Enhanced Binary Particle swarm Optimization (EBPSO), (ii) To mine data using various data chunks (windows) and overcome a failure of single clustering. An experimental analysis is conducted by several experiments to assess the efficiency of the suggested methods that have been tested within the benchmark datasets, namely CICIDS2017. In comparison to different metaheuristic algorithms for feature selection, experimental outcomes indicate that the suggested method is capable of reducing dimensionality cost, the number of irrelevant features and produce reasonable accuracy. Experiments demonstrate and prove that the proposed EBPSO method produces better accuracy mining data and selecting subset of relevant features comparing other algorithms. In addition, experiments prove that the enhanced algorithm shows a higher performance through lower false positive, higher accuracy, and better CPU time. 2019 Thesis https://ir.uitm.edu.my/id/eprint/83493/ https://ir.uitm.edu.my/id/eprint/83493/1/83493.pdf text en public phd doctoral Universiti Teknologi MARA (UiTM) Faculty of Electrical Engineering Hashim, Habibah |