Designing an Efficient Network-Based Intrusion Detection System Using an Artificial Bee Colony and ADASYN Oversampling Approach
Manisha Rani, Gunreet Kaur, Gagandeep Gagandeep · 2022
Intrusion detection system (IDS) is the fundamental mechanism to secure network packets from suspicious activities in order to ensure integrity and availability of the network resources. Due to the large dimensional size of IDS datasets, it has from a high false-alarm rate and low testing accuracy when separating attack data from normal data. Besides this, the class imbalance problem is also a critical issue, and greatly affects the performance of the IDS model. In order to reduce the dimensions of the dataset and to address class imbalance issue, we have used the Artificial Bee Colony (ABC) algorithm for feature selection and the Adaptive Synthetic (ADASYN) oversampling technique to balance the dataset. Initially, data is preprocessed through data conversion and the min-max normalization method. Subsequently, data is balanced by oversampling the minority instances to equalize them to majority instances in the dataset. The optimal feature subset is chosen using the ABC algorithm, followed by binary classification using Random forest classifier. The experimental results are calculated over the well-known NSL KDD dataset and the performance of proposed work is compared with existing literature in order to validate the results.