Energy Efficiency Data Mining for Wireless Sensor Networks Based on Random Forests

In this paper, we propose a novel data mining technique involving random forests and random trees for energy efficiency for forest cover type classification. Novel machine learning and data mining techniques provide an unprecedented opportunity to monitor and characterize physical environments, such as forest cover type, using low cost wireless sensor networks. However, given the sheer amount of data collected by the wireless sensor networks, conventional classification schemes that demand enormous storage and computation requirements make them less effective for deployment in real world application scenarios. There is a need for intelligent and energy efficient monitoring techniques, made possible by novel data mining and classification techniques, and the work reported in this paper involves such a novel energy efficient data mining scheme for forest cover type classification based on random forests and random trees. The experimental validation of the proposed data mining scheme on a publicly available UCI machine learning dataset, shows that the proposed random forest and random tree based approach perform significantly better than the conventional statistical classifiers, such as Naïve Bayes, discriminant classifiers and decision trees, and can lead towards energy efficient, intelligent monitoring and characterization of large physical environments instrumented using wireless sensor networks.