BDD: Efficient and Scalable Collection, Analytics and Processing of Big Data for Disaster Applications

Missouri University Of Science And Technology, Rolla MO

Investigators

Abstract

The outcomes from this project is to assist human operators in their disaster management coordination and planning, such as directing a medical physician's team to their nearest cluster of affected people in a region to administer medications as necessary, or finding a safe route for evacuation of affected people. Sensor data integrated with microblogs such as Tweets help identifying some local events and people's sentiments, which are significantly useful in handling/understanding disaster situations. It will also benefit other applications such as real-time tracking of road/driving conditions in vehicular networks. This research is conducted jointly with Osaka University in Japan, to benefit both the universities in enhancing not only their knowledge but also to learn global perspective in solving important problems. The research team is designing schemes for dynamic and collaborative data compression and multi-streams compression of multi-dimensional sensor data with error correction and recovery for addressing the energy efficiency and bandwidth limitation issues. Compression schemes exploit temporal locality and delta compression to provide better bandwidth utilization. Different methods for measuring error are designed and compared for the compressibility and actual error for variations in methods of utilizing the error tolerance. In addition, the team is developing algorithms for highly scalable indexing schemes for efficient data retrieval involving mainly range queries, top-k query, ranked-based searches and snapshot queries for multi-dimensional sensor data from different data sources to address the issue of timely dissemination. Hilbert Curve based linearization technique integrated with an overlay network is designed to (1) map multidimensional attributes onto a single dimension while preserving its data locality, and (2) to create a balanced network by associating only one node with each leaf of the virtual tree and then partition the multidimensional search space into subspaces and assign each node to a unique subspace. This allows an overlay network to start from a predefined prefix to handle data skewness. This research is also designing a scheme for using microblog messages as social sensors for efficient integration with other sensor data. We are using machine-learning techniques to match each message with its associate location based on the characteristics of the message. The results will be validated and evaluated using the sensor cloud test-bed available at Missouri S&T.