Optimal intrusion detection for imbalanced data using Bagging method with deep neural network optimized by flower...

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning

Background:

• The proliferation of Internet of Things (IoT) devices necessitates robust security measures.
• Intrusion Detection Systems (IDS) are crucial for identifying malicious activities in IoT networks.
• Class imbalance in intrusion datasets poses a significant challenge for IDS development.

Purpose of the Study:

• To propose a novel hybrid IDS for IoT networks.
• To enhance intrusion detection accuracy and handle class imbalance.
• To leverage metaheuristic and deep learning techniques for improved network security.

Main Methods:

• A hybrid approach combining the Flower Pollination Algorithm (FPA) and Deep Neural Networks (DNN).
• An ensemble learning paradigm utilizing a roughly-balanced (RB) Bagging strategy.
• FPA-trained DNNs with a cost-sensitive fitness function as base learners for unbiased model training.

Main Results:

• The proposed IDS demonstrated superior performance across four benchmark datasets (NSL-KDD, UNSW NB-15, CIC-IDS-2017, BoT-IoT).
• Effective handling of class imbalance was achieved through the RB Bagging strategy.
• High accuracy, precision, recall, and F1-score were reported, outperforming existing IDS.

Conclusions:

• The hybrid FPA-DNN IDS offers an effective solution for detecting intrusions in IoT environments.
• The RB Bagging strategy successfully mitigates the challenge of imbalanced datasets.
• This approach provides a significant advancement in securing IoT networks against cyber threats.