Enhancing Intrusion Detection Systems with Dimensionality Reduction and Multi-Stacking Ensemble Techniques

Algorithms, Год журнала: 2024, Номер 17(12), С. 550 - 550

Опубликована: Дек. 3, 2024

The deployment of intrusion detection systems (IDSs) is essential for protecting network resources and infrastructure against malicious threats. Despite the wide use various machine learning methods in IDSs, such often struggle to achieve optimal performance. key challenges include curse dimensionality, which significantly impacts IDS efficacy, limited effectiveness singular classifiers handling complex, imbalanced, multi-categorical traffic datasets. To overcome these limitations, this paper presents an innovative approach that integrates dimensionality reduction stacking ensemble techniques. We employ LogitBoost algorithm with XGBRegressor feature selection, complemented by a Residual Network (ResNet) deep model extraction. Furthermore, we introduce multi-stacking (MSE), novel method, enhance attack prediction capabilities. evaluation on benchmark datasets as CICIDS2017 UNSW-NB15 demonstrates our surpasses current models across performance metrics.

Язык: Английский

---

An Integrated Approach for Intrusion Detection in Intelligent Grid Computing Networks Using Machine Learning

Journal of Wireless Mobile Networks Ubiquitous Computing and Dependable Applications, Год журнала: 2024, Номер 15(4), С. 313 - 324

Опубликована: Дек. 12, 2024

Intelligent Grid (IG) systems improve the usability of old energy networks, but they can still be hacked in many ways. Intruders get into system through these holes, risking IG networks' safety and privacy. An Intrusion Detection System (IDS) keeps services safe secure an setting. With help Machine Learning (ML) techniques characteristics, this work shows IDS for platforms. The categorization algorithm comprises a Convolutional Neural Network (CNN) Gated Recurrent Unit (GRU). research uses Precision, Detecting Rate (IDR), False Alarming Ratio (FAR) to rate how well suggested approach works. It turns out that Random Forest (RF) (NN) algorithms did outperform others. study found KDD-99 records had Alarm 7.29%, NSL-KDD FAR 7.31%. 88.68% time, both methods find things, 90.87% confirm are correct.

Язык: Английский

---

AI for Cybersecurity: Robust models for Authentication, Threat and Anomaly Detection

MDPI eBooks, Год журнала: 2023, Номер unknown

Опубликована: Июль 21, 2023

compare different ensemble learning methods that have been proposed in this context: Random Forests, XGBoost, CatBoost, GBM, and LightGBM.Experiments were performed on datasets, finding tree-based algorithms can achieve good performance with limited variability. Access Control [7,8]As stated above, access control be viewed as another point the anomaly detection continuum.Again, distinguishing a legitimate user from impostors automated through machine learning.The seventh paper [7] addresses context of face recognition systems (FRSs) proposes practical white box adversarial attack algorithm.The method is evaluated CASIA WebFace LFW datasets.In [8], authors used user's iris image, combined secret key, to generate public key subsequently use such data limit protected resources. Threat Intelligence [9,10]Not only do we want recognize block attacks they occur-we also need observe external overall network predict relevant events new patterns, addressing so-called threat intelligence landscape.In [9], two well-known databases (CVE MITRE) technique link correlate these sources.The tenth [10] formal ontologies monitor threats identify corresponding risks an way. ConclusionsIn conclusion, observed AI increasingly being cybersecurity, three main directions current research: (1) areas cybersecurity are addressed, CPS security intelligence; (2) more stable consistent results presented, sometimes surprising accuracy effectiveness; (3) presence AI-aware adversary recognized analyzed, producing robust reliable solutions.

Язык: Английский

---