Mobile-Agent-Based Distributed Decision Tree Classification in Wireless Sensor Networks

Wireless sensor networks (WSNs) connect physical sensors that are distributed in the environment. In many applications, the statistical pattern recognition methods, such as decision tree (DT) algorithms, are used to recognize the patterns of the sensor readings. To enable the decision tree classification (DTC) in WSNs, a new distributed decision tree classification (DDTC) algorithm based on mobile agent is proposed in this paper. We organize the conjunctive sets of linear classifiers in DT into groups of operations on attributes. Each group of operations is allocated to a single sensor node. If a mobile agent visits these sensor nodes serially, the recognition result can be acquired step by step. Thus the sensor nodes do not need to transmit all the sensor data to a centralized node where all the data processing is carried out in traditional DTC. In DDTC, if not all the attributes are needed for operation, classification of one instance can finish halfway. Two public data sets are used to evaluate the performance of DTC and DDTC on energy consumption. The simulation results indicate that, compared with the centralized DTC, the DDTC algorithm decreases the number of transmissions, balances the power consumption and computation among sensor nodes, and prolongs the lifetime of the network.