Standard classi??cation algorithms are o?en inaccurate when used in a wireless sensor network (WSN), where the observed data\r\noccur in imbalanced classes. ?e imbalanced data classi??cation problem occurs when the number of samples in one class, usually\r\nthe class of interest, is much lower than the number in the other classes. Many classi??cation models have been studied in the datamining\r\nresearch community. However, they all assume that the input data are stationary and bounded in size, so that resampling\r\ntechniques and postad??ustment by measuring the classi??cation cost can be applied. In this paper, we devise a new scheme that\r\nextends a popular stream classi??cation algorithm to the analysis of WSNs for reducing the adverse effects of the imbalanced class\r\nin the data. ?is new scheme is resource light at the algorithm level and does not require any data preprocessing. It uses weighted\r\nna�¯ve Bayes predictors at the decision tree leaves to effectively reduce the impact of imbalanced classes. Experiments show that our\r\nmodi??ed algorithm outperforms the original stream classi??cation algorithm.
Loading....