IoT Equipment Monitoring System Based on C5.0 Decision Tree and Time-Series Analysis

Biaokai Zhu, Xinyi Hou, Sanman Liu, Wanli Ma, Meiya Dong, Haibin Wen, Qing Wei, Sixuan Du, Yufeng ZhangIEEE Access 2022 · DOI

Introduction: The Growing IoT Security Gap

The Internet of Things (IoT) is expanding at a breakneck pace, integrating into everything from smart homes to industrial infrastructure. However, this rapid deployment has outpaced the development of robust security frameworks. Traditional traffic monitoring systems often struggle with the sheer volume and variety of IoT data, frequently relying on manual feature engineering and static rules that fail against "zero-day" or unknown attacks.

Common attack vectors in an IoT network environment.

Researchers identify three primary challenges in current IoT security:

  1. High Manual Overhead: Classifying traffic and updating threat signatures requires extensive human intervention.
  2. Poor Adaptability: Static models cannot easily adjust to the evolving nature of network threats.
  3. The "Unknown" Factor: Most systems are designed to recognize known signatures, leaving them vulnerable to novel exploits.

Proposed Multi-Layered Architecture

To address these issues, a hybrid monitoring system has been developed that combines the interpretability of decision trees with the predictive power of deep learning. The architecture treats network security as both a classification problem and a computer vision task.

The proposed hybrid monitoring system architecture.

Core Components

  • C5.0 Decision Tree: This serves as the primary classification engine. Chosen for its efficiency in handling large datasets, the C5.0 algorithm uses information entropy to split data into informative subsets, distinguishing normal traffic from malicious patterns.
  • Gramian Angular Field (GAF): This technique transforms one-dimensional time-series traffic data into two-dimensional images. By "visualizing" traffic, the system preserves temporal correlations in a spatial format, allowing the use of advanced image recognition models.
  • CNN-LSTM Hybrid Model: The GAF images are processed by a Convolutional Neural Network (CNN) to extract local features. These features are then passed to a Long Short-Term Memory (LSTM) network, which excels at understanding long-term dependencies and trends over time, crucial for detecting slow-acting or multi-stage attacks.
  • Whitelist Matching: To ensure results are actionable, the system uses whitelist technology to match data patterns to specific physical IoT device models.

Datasets and Evaluation

The study utilized the KDD Cup 99 dataset, a classic benchmark in intrusion detection. The researchers used a randomly selected 10% subset of the 5 million records, covering major attack types such as Denial of Service (DoS), User to Root (U2R), Remote to Local (R2L), and Probing. To supplement this, real-world traffic was collected from 43 types of intelligent devices via the Device Hive IoT virtual platform.

Key Results

The hybrid approach yielded impressive quantitative metrics:

  • Overall Accuracy: The system achieved a detection accuracy of 96.71%.
  • Unknown Attack Detection: It successfully identified previously unseen attacks with 96% accuracy.
  • Low Error Rates: The average error rate for abnormal traffic was just 3.22%.
  • Specific Performance: DoS attacks were detected with 96.14% accuracy, while Probing attacks reached 97.98%.

Limitations and Future Outlook

While the system demonstrates high efficacy, scaling the data-to-equipment matching system across the vast heterogeneity of the IoT market remains a challenge. Furthermore, while KDD Cup 99 is a standard benchmark, its age suggests that future testing against modern datasets like UNSW-NB15 or CICIDS2017 is necessary to validate performance against contemporary threats. Despite these hurdles, the integration of GAF-based visualization and hybrid deep learning represents a significant step forward in automated, adaptive IoT security.

Summary updated

BibTeX
@article{zhu2022iot,
  author = {Biaokai Zhu and Xinyi Hou and Sanman Liu and Wanli Ma and Meiya Dong and Haibin Wen and Qing Wei and Sixuan Du and Yufeng Zhang},
  title = {{IoT Equipment Monitoring System Based on C5.0 Decision Tree and Time-Series Analysis}},
  journal = {IEEE Access},
  year = {2022},
  doi = {10.1109/access.2021.3054044},
}
Figure