A New Representation Structure for Mining Association Rules from Wireless Sensor Networks

With the advances in wireless sensor networks and their ability to generate a large amount of data, data mining techniques used to extract useful knowledge regarding the underlying network have recently received a great deal of attention. In this paper the authors use a new formulation for association rules, a well known data mining technique. This formulation allow us to extract the associations between sensors, which will make it easy to predict the set of sensors that can report events in the same time interval. In order to generate these rules, frequent patterns should be determined. A pattern is a subset of sensors and is frequent if it occurs a certain number of times. However, the stream nature and the large amount of data, in addition to the big number of possible patterns, add more challenges to the traditional data mining algorithms developed to generate the association rules. To overcome these limitations, the authors propose a new representation structure for the sensor data. This structure, which they call positional lexicographic tree (PLT), is a able to partition the data and present them in a sorted and compressed format, and provides an easy mechanism to access and manipulate the data. Also, the authors propose a data mining algorithm that follows a pattern growth approach to extract the frequent patterns efficiently. The authors compare the performance of the mining algorithm with FP-Growth, a well known algorithm in pattern growth approach. The results have shown that PLT structure and its mining routine outperforms the FP-Growth in both CPU time and memory usage.