Abstract
Recently, many data mining techniques have been applied to analyze and interpret the huge volume of data collected from wireless sensor networks. Such techniques, especially classification and clustering, have been used to relate raw data and assign a class label (a useful interpretation) to the set of attributes values received from the sensors' nodes. However, building a classifier, such as decision tree, is a cost process in terms of energy consumption due to the large size of the resultant tree. In this article, we propose a model-checking based classification method that relies on cutting-off parts of the decision tree while keeping the performance fixed. The pruning process aims to reduce the size of the tree and, thus, reduce the amount of the energy needed to maintain the classifier. Results have shown energy reduction between 10–15% compared with a nonpruned decision tree.