turboBurst: A High Dimensional Data Classification approach for identifying Bursty Links in a Highly Spatiotemporal Correlated Sensor Network

Wireless Sensor Network (WSN) exhibits dynamic link behaviors. Choosing a stable route is essential for a reliable end-to-end communication. Link identification is important for utilization of both long and short terms high quality links for packet forwarding. Machine Learning has proved to learn wireless link dynamics with precision and accurately identify stable short-term link quality. This paper focuses on a data-driven design framework, turboBurst, that identifies bursty links in a highly spatiotemporal correlated Sensor Network. A metric PDD is used to measure the degree of link burstiness using a Two State Markov Chain and the Earth Mover's Distance approaches as our Conditional Probability Distribution model. The model and a Kernelized Linear Support Vector Machine are exploited to identify numerous bursty links at various window sizes. The efficiency of our technique is studied over a data set obtained from an 802.15.4 network. These approaches accurately estimates between 40 and 47 short-term high quality links in the spatial temporal correlated network. Thus turboBurst shows a good performance on quality short-term link identification accuracy.