LAUSR

Abstract Early prediction of a forest fire is one of the critical research challenges of the wireless sensor network (WSN) to save our ecosystem. In WSN based forest fire detection system, sensor nodes are deployed in the remote forest area for transmitting the sensed data to the base station, which is accessible by the forest department. Though sensor nodes in the forest are localized through GPS connection, the high deployment cost for it motivates the authors of this paper to design a novel localization technique applying the Support Vector Machine. Forest fire prediction in an energy efficient way is another concern of this paper. The semi-supervised classification model is proposed to address this problem by dividing the forest area into different zones [High Active (HA), Medium Active (MA), and Low Active (LA)]. It is designed in such a way that it can be able to predict the state of the (HA, MA, LA) fire zone with 90% accuracy when only one parameter is sensed by sensor nodes due to energy constraints. The greedy forwarding technique is used to transmit the packets from the HA zone to the base station continuously, and the MA zone transmits packets periodically, whereas, LA zone avoids transmitting the sensed data to the base station. This technique of data forwarding enhances network lifetime and reduces congestion during data transmission from the forest area to the base station. Graphic abstract