Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Optimized Ensemble Model Towards Secured Industrial IoT Devices

MohammadNoor Injadat

TL;DR

The paper addresses intrusion detection for Industrial IoT by proposing a framework that couples Bayesian Optimization-Gaussian Process with an ensemble tree learner to optimize hyper-parameters and reduce overfitting. It uses a Windows 10 Ton IoT dataset and data-preprocessing steps to handle missing values and scale features, demonstrating improvements in accuracy, precision, and F-score over standard tree models. The key contributions include a data-preprocessing pipeline, BO-GP-driven hyper-parameter optimization, and empirical evidence that the optimized ensemble outperforms comparable models, with faster convergence. This approach advances IIoT security by providing a more accurate and efficient intrusion-detection framework suitable for resource-constrained industrial settings.

Abstract

The continued growth in the deployment of Internet-of-Things (IoT) devices has been fueled by the increased connectivity demand, particularly in industrial environments. However, this has led to an increase in the number of network related attacks due to the increased number of potential attack surfaces. Industrial IoT (IIoT) devices are prone to various network related attacks that can have severe consequences on the manufacturing process as well as on the safety of the workers in the manufacturing plant. One promising solution that has emerged in recent years for attack detection is Machine learning (ML). More specifically, ensemble learning models have shown great promise in improving the performance of the underlying ML models. Accordingly, this paper proposes a framework based on the combined use of Bayesian Optimization-Gaussian Process (BO-GP) with an ensemble tree-based learning model to improve the performance of intrusion and attack detection in IIoT environments. The proposed framework's performance is evaluated using the Windows 10 dataset collected by the Cyber Range and IoT labs at University of New South Wales. Experimental results illustrate the improvement in detection accuracy, precision, and F-score when compared to standard tree and ensemble tree models.

Optimized Ensemble Model Towards Secured Industrial IoT Devices

TL;DR

The paper addresses intrusion detection for Industrial IoT by proposing a framework that couples Bayesian Optimization-Gaussian Process with an ensemble tree learner to optimize hyper-parameters and reduce overfitting. It uses a Windows 10 Ton IoT dataset and data-preprocessing steps to handle missing values and scale features, demonstrating improvements in accuracy, precision, and F-score over standard tree models. The key contributions include a data-preprocessing pipeline, BO-GP-driven hyper-parameter optimization, and empirical evidence that the optimized ensemble outperforms comparable models, with faster convergence. This approach advances IIoT security by providing a more accurate and efficient intrusion-detection framework suitable for resource-constrained industrial settings.

Abstract

The continued growth in the deployment of Internet-of-Things (IoT) devices has been fueled by the increased connectivity demand, particularly in industrial environments. However, this has led to an increase in the number of network related attacks due to the increased number of potential attack surfaces. Industrial IoT (IIoT) devices are prone to various network related attacks that can have severe consequences on the manufacturing process as well as on the safety of the workers in the manufacturing plant. One promising solution that has emerged in recent years for attack detection is Machine learning (ML). More specifically, ensemble learning models have shown great promise in improving the performance of the underlying ML models. Accordingly, this paper proposes a framework based on the combined use of Bayesian Optimization-Gaussian Process (BO-GP) with an ensemble tree-based learning model to improve the performance of intrusion and attack detection in IIoT environments. The proposed framework's performance is evaluated using the Windows 10 dataset collected by the Cyber Range and IoT labs at University of New South Wales. Experimental results illustrate the improvement in detection accuracy, precision, and F-score when compared to standard tree and ensemble tree models.
Paper Structure (13 sections, 4 equations, 3 figures, 1 table)

This paper contains 13 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. General ML-based Network Attack Detection Works:
  4. ML-based IoT Network Attack Detection Works:
  5. Proposed Approach
  6. Description:
  7. Computational Complexity:
  8. Dataset Description
  9. Experiment Results & Discussion
  10. Experiment Setup
  11. Performance Metrics
  12. Proposed Framework's Performance Results & Discussion
  13. Conclusion & Future Works

Figures (3)

  • Figure 1: Proposed Optimized Ensemble Learning Framework for Intrusion Detection in IIoT Environments
  • Figure 2: First and Second Principal Components of Windows 10 Dataset
  • Figure 3: Minimum Classification Error Progression During Optimization