A Study on Feature Analysis and Ensemble-based Intrusion Detection Scheme using CICIDS-2017 dataset
- Publication Type:
- Thesis
- Issue Date:
- 2021
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
One of the primary security research challenges faced by traditional IDS methods is their inability to handle large volumes of network data and detect modern cyber-attacks with high detection accuracy and low false alarms. Hence, there is a need for efficient and reliable IDS schemes that can tackle this ever-changing cybersecurity paradigm. Machine learning techniques are hence, becoming very popular in designing modern intrusion detection systems. Several supervised and unsupervised machine learning techniques have been used in literature; however, the IDS classification efficiency is affected by noisy data in high dimensional datasets. The role of feature selection is significant as the feature selection process eliminates the redundant and noisy data and further selecting optimal feature subset enables reduction of high dimensional IDS datasets. Machine learning algorithms are extensively being used for intrusion detection. However, research has proved that the performance of multiple classifier-based IDS is far better than an IDS classifier, which has given us the motivation to develop an ensemble-based intrusion detection model. Lastly, the benchmark IDS datasets currently being used for the evaluation of IDS schemes are outdated and do not represent modern-day attacks. The CICIDS-2017 dataset is offered by the University of New Brunswick. It is the latest publicly available dataset for intrusion detection. However, there are a significantly low number of research studies conducted using this dataset which also focus on optimal feature selection. This dataset has a good potential to be used as a future benchmark intrusion detection dataset as it covers the modern-day system setup and threat profile and the dependency on outdated IDS datasets can be removed. There is a need to benchmark the performance of modern IDS datasets using machine learning ensemble-based classifiers. This thesis aims to address the issues by proposing a new intrusion detection framework using ensemble-based feature selection method for generating a low dimensionality feature subset and ensemble-based intrusion detection framework to benchmark the performance of the CICIDS - 2017 dataset. The proposed scheme is beneficial for research community as it combines the use of the latest available IDS dataset with ensemble technique for feature selection and ensemble-based intrusion detection model.
Please use this identifier to cite or link to this item: