With enormous use of sensor network, the sensor applications was enfeeble by large data movements across the network owing to the intrinsic characteristics of resource-constrained sensors. This paper emphasizes on two issues such as accuracy and communication overhead, data aggregation is an effective way to reduce communication issues at lower level, to take important decisions and accuracy at final aggregation result. Next a model is predict from the extracted data using supervised learning at the upper level thus reduce the communication effort, the high dimensional reduction of the data mining task, then improving the accuracy of the sensing task.