Data mining approach to situation-aware sensor actuation in wireless sensor networks

As wireless sensor networks (WSN) are used in seemingly limitless applications with more sensor types available, large sensor data sets are acquired and transmitted for additional processing. These large data sets may cause network traffics slow unnecessarily. The unfiltered transmission of raw data will also consume the WSN batteries inefficiently. In recent years, a few techniques have been developed for event-driven WSNs and sensor collaborations. For example, in traffic surveillance camera systems, video clips are transmitted in the event of car collision noises. Event-driven services run as a preset parameter of surveillance sensing. It is likely that there may be a false-positive event which drives video data to be transmitted unnecessarily, or false-negative event which misses alarms, both leading to accurate surveillance. This paper proposes an adaptive technique for event-driven WSNs, in which learning from a classification process is employed to fine-tune pre-defined data-driven WSNs. Unlike to traditional data classification techniques, outliers of collected sensor data will be one of the critical data points. The proposed technique functions more proactively by being aware of the environmental situation changes. The contribution of this paper includes an adaptive technique of acquiring and utilizing sensor data.