Resource-Aware Time Series Imaging Classification for Wireless Link Layer Anomalies.

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• The proliferation of smart infrastructure drives a dramatic increase in end devices relying on last-mile wireless connectivity.
• Efficient management of these massive wireless networks necessitates advanced solutions for monitoring and malfunction detection.

Purpose of the Study:

• To analyze image-based representation techniques for wireless anomaly detection.
• To propose a novel, resource-aware deep learning architecture for accurate wireless anomaly detection.

Main Methods:

• Utilizing recurrence plots (RPs) for time series to image transformation.
• Developing and evaluating a new deep learning architecture for anomaly detection.
• Comparing the proposed model against Gramian angular fields, classical machine learning models, and mainstream deep learning architectures.

Main Results:

• The proposed RP-based model outperforms Gramian angular fields by up to 14% and dynamic time warping by up to 24%.
• The model achieves comparable or superior performance to AlexNet and VGG11 with significantly lower computational complexity (≈8%).
• The new architecture surpasses the state-of-the-art in wireless anomaly detection by up to 55%.

Conclusions:

• Image-based representation using RPs offers a highly effective approach for wireless anomaly detection.
• The proposed deep learning architecture provides an accurate, efficient, and resource-aware solution for monitoring massive wireless networks.
• This work advances the field of network monitoring, enabling more reliable smart infrastructure operations.