Cleaning of Data Streams in Mobile Wireless Sensor Networks

Speaker Bio

Samia Tasnim is a Ph.D. candidate at the School of Computing and Information Sciences (SCIS), Florida International University (FIU) under the supervision of Dr. S. S. Iyengar and Dr. Niki Pissinou. She received her Bachelor’s degree in Computer Science and Engineering from Bangladesh University of Engineering and Technology (BUET) and Master’s degree in Computer Science from FIU in 2015. Her research interests include mobile systems, wireless sensor networks, database and data mining. Samia worked as a software engineer for two years before starting graduate studies at FIU.

Description

Mobile Wireless Sensor Networks (MWSNs) are a critically important technology area for the acquisition of spatiotemporally dense data in diverse applications, ranging from environmental monitoring to intrusion detection and surveillance. However, due to their depletion of energy resources, deployment in hazardous environments, mechanical faults or malicious attacks by adversaries, sensor data is subject to several sources of errors such as noise, inaccuracies, and imprecision. Existing data cleaning techniques specifically developed for the MWSNs, focus on using post-processing data cleaning at the server end. These methods cannot ensure real-time data accuracy, as they process data in batches after long intervals of time, and thus cannot take immediate action when required. On the other hand, context-awareness allows us to examine the computing environment and react to environmental changes. Recently, scarce research has focused on geographical context of the sensors. Since contextual relationship was only incorporated with either spatial model or time series model, researchers were unable to utilize the relationship among the sensors to the fullest. While researchers have attempted to improve the quality of the received sensor data, no work has been done on how sensor context (e.g., terrain elevation, wind speed, user movement during sensing) can be used along with spatiotemporal relationship in correlated sensor selection for online data cleaning.

This dissertation aims at developing online methods by fusing spatiotemporal and context relationships among the participating mobile sensors. To identify contextual relationships that can be used to improve in-network, real-time cleaning of data streams, it is necessary to find if contextual relationship along with spatiotemporal relationship will improve cleaning of data streams in real-time. To do so, we developed a novel data cleaning mechanism where, based on the sensed data and the context relationship of each sensor, we update the credibility of the sensed data. Preliminary results indicate promising outcomes to this proposed method through achieving higher accuracy than state-of-the-art methods. The prediction error that our method achieves is less than those from related contemporary works. Second, we propose a window-based online sub-trajectory clustering method for finding movement similarity based on space, time, direction, and semantics. Preliminary results using a real-world application dataset indicate promising outcomes to this proposed semantic-aware trajectory data mining method.

Our method does not perform well in sparse networks where the concurrent presence of more than one sensor in close proximity is not ensured. In the future, we propose to investigate the different density level of networks to improve data stream cleaning in sparse networks. Moreover, in mobile crowd sensing, no user can be trusted as people tend to provide more amount of data for financial gain, but may ignore the data quality. We will further develop an online method for data quality prediction in mobile crowd sensing that considers heterogeneous trust level of the participating users.