Due to the increased usage of the Internet of Things and heterogeneous distributed devices, the development of effective and reliable intrusion detection systems (IDS) has become more critical. The massive volume of data with various dimensions and security features, on the other hand, can influence detection accuracy and raise the computation complexity of these systems. Fortunately, Artificial Intelligence (AI) has recently attracted a lot of attention, and it is now a principal component of these systems. This work presents an enhanced intelligent intrusion detection model (E2IDS) to detect state of the art known cyberattacks. The model design is Decision Tree (DT) algorithm-based, with an approach to data balancing since the data set used is highly unbalanced and one more approach for feature selection. Furthermore, accuracy, recall and F-score are selected as the performance evaluation metrics. The experimental results show that our E2IDS not only overcomes the benchmark work but also reduces the complexity of the computing process.
30 October 2024 in International Journal of Information Security
Feature interaction is a vital aspect of Machine Learning (ML) algorithms, and gaining a deep understanding of these interactions can significantly enhance model performance. This paper introduces the BukaGini algorithm, an innovative and robust approach for feature interaction analysis that capitalizes on the Gini impurity index. By exploiting the unique properties of the BukaGini index, our proposed algorithm effectively captures both linear and nonlinear feature interactions, providing a richer and more comprehensive representation of the underlying data. We thoroughly evaluate the BukaGini algorithm against traditional Gini index-based methods on various real-world datasets. These datasets include the High School Students’ Performance (HSSP) dataset, which examines factors affecting student performance; Cancer Data, which focuses on identifying cancer types based on gene expression; Spambase, which targets spam email classification; and the UNSW-NB15 dataset, which addresses network intrusion detection. Our experimental results demonstrate that the BukaGini algorithm consistently outperforms traditional Gini index-based methods in terms of accuracy. Across the tested datasets, the BukaGini algorithm achieves improvements ranging from 0.32% to 2.50%, underscoring its effectiveness in handling diverse data types and problem domains. This performance gain highlights the potential of the BukaGini algorithm as a valuable tool for feature interaction analysis in various ML applications.