Securing IoT Networks: A Machine Learning Approach for Detecting Unusual Traffic Patterns

Overview

Securing IoT Networks: A Machine Learning Approach for Detecting Unusual Traffic Patterns is a Scientific Reports article that presents a unified intrusion detection framework by merging two complementary IoT security datasets and applying machine learning algorithms to detect both IoT-specific and general network threats. The study reports high detection performance using various classifiers.

Dataset Composition

• N-BaIoT dataset: network traffic from commercial IoT devices infected with Mirai and BASHLITE botnets.
• UNSW-NB15 dataset: a modern network intrusion dataset with multiple attack families and normal traffic.
• Integration approach: traffic from both datasets is merged with aligned features and labels to form a unified training corpus.

Feature Engineering and Optimization

• Flow-based network statistics such as packet counts, byte rates, inter-arrival times, flags, ports, and protocol distributions are extracted.
• Correlation analysis, mutual information, and Recursive Feature Elimination are used to identify the most discriminative features.
• Dimensionality reduction reduces feature space while maintaining detection performance.
• Data normalization and class balancing techniques (e.g., SMOTE) are applied to handle scale differences and class imbalance.

Machine Learning Models

• Evaluated models include Decision Tree, SVM, Random Forest, and Neural Network classifiers.
• Neural Networks achieve the highest accuracy and strong precision/recall across attack classes.
• Models are evaluated with cross-validation and separate test sets to ensure robustness.

Use Cases

• Hybrid IoT–IT network security where IoT devices coexist with traditional IT infrastructure.
• Real-time anomaly detection on gateways or in the cloud.
• Benchmarking new ML and DL algorithms against established baselines.

Access and License

The article describes the merged dataset and processing methodology using publicly available N-BaIoT and UNSW-NB15 datasets. Access follows Scientific Reports open-access policies and the licenses of the underlying datasets.