Artificial Intelligence-Driven Intrusion Detection in Software-Defined Wireless Sensor Networks: Towards Secure...

Area of Science:

• Computer Science
• Network Security
• Artificial Intelligence

Background:

• Wireless Sensor Networks (WSNs) are crucial for Internet of Things (IoT) applications, handling vast amounts of critical data.
• Efficient management and security are paramount for WSNs, leading to the development of Software-Defined Wireless Sensor Networks (SDWSNs).
• Intrusion Detection Systems (IDS) are vital for safeguarding SDWSN-based IoT environments.

Purpose of the Study:

• To evaluate the effectiveness of three Artificial Intelligence (AI) techniques—Decision Tree, Naïve Bayes, and Deep Artificial Neural Network—as anomaly detectors in IDSs for SDWSNs.
• To compare the performance metrics of these AI models for both binary and multinomial intrusion classification.
• To introduce an advanced feature engineering scheme to enhance detection accuracy.

Main Methods:

• Trained Decision Tree, Naïve Bayes, and Deep Artificial Neural Network models on the NSL-KDD dataset.
• Implemented an end-to-end feature engineering process to expand the dataset from 41 to 118 features.
• Evaluated models based on accuracy (binary classification) and F-scores (multinomial classification).

Main Results:

• The Decision Tree-based IDS demonstrated superior performance, achieving state-of-the-art accuracy of 0.999777 in binary classification and high F-scores in multinomial classification.
• The Naïve Bayes model proved suitable for binary classification in resource-constrained SDWSN-IoT devices.
• The Deep Artificial Neural Network shows significant promise as a future default anomaly detector due to its current performance and the growing availability of training data.

Conclusions:

• Decision Tree-based anomaly detection is highly effective for securing SDWSN-based IoT systems.
• Naïve Bayes is a viable option for low-memory IoT devices requiring binary intrusion detection.
• Deep Artificial Neural Networks are poised to become the standard for anomaly detection in WSNs as data availability increases.