A Comparative Study of Anomaly Detection Techniques for Smart City Wireless Sensor Networks

Area of Science:

• Computer Science
• Urban Planning
• Data Science

Background:

• Smart cities leverage wireless sensor networks (WSN) and the Internet of Things (IoT) to enhance urban services and operational efficiency.
• Data integrity in smart cities is crucial but challenged by WSN errors and sophisticated cyber and physical attacks.
• Existing anomaly detection methods require evaluation for their efficacy in securing smart city data streams.

Purpose of the Study:

• To simulate WSNs and implement common attacks using real data from Barcelona, a leading smart city.
• To compare the performance of frequently used anomaly detection techniques in identifying these simulated attacks.
• To assess algorithm effectiveness under varying network status information constraints.

Main Methods:

• Utilized real-world smart city data from Barcelona for WSN simulations.
• Implemented typical physical and computer-based attacks targeting WSN data integrity.
• Compared multiple anomaly detection algorithms, including one-class Support Vector Machines (SVM).
• Evaluated algorithm performance based on true positive and false positive rates under different network information scenarios.

Main Results:

• One-class SVM demonstrated superior performance in detecting WSN attacks compared to other methods.
• Achieved at least a 56% higher true positive rate with a 5% false positive rate.
• Secured a 26% higher true positive rate in a scenario with a 15% false positive rate.

Conclusions:

• One-class SVM is the most suitable anomaly detection technique for securing smart city WSN data.
• The findings provide a robust method for enhancing the reliability and security of smart city infrastructure.
• Effective anomaly detection is critical for maintaining the integrity and trustworthiness of smart city services.