The Internet of things concept is based on data from sensor devices. The conservation of sensor data in the cloud has become a well-known challenge with respect to the sensor-cloud network. There is an enormous quantity of sensed data, including missed and unimportant data due to unavailability of the node, low link quality, node deficiency or latency in path communication. In order to address this issue, it is necessary to develop an accurate sensor selection approach, precisely for the purpose of reducing the quantity of unimportant data (missed and noisy data). In this paper, we formulate an innovative and efficient bi-level node energy assessment algorithm, named BNSA, which uses a belief propagation protocol for optimal energy-efficient sensor selection. On the first level, the equilibrium transmission energy function is applied to all nodes to evaluate the transmission rate and link quality, together with the residual energy of the node, to predict the energy consumption, providing an accurate prediction of the energy level of the nodes. On the second level, optimal nodes are selected based on the prediction of the energy level. Belief propagation draws inferences for revealing missed sensor data. This approach improves the lifespan of the network and reduces the energy utilization rate by more than 32.3% with a delay rate of 77%. This is reflected in an enhancement of the quality of service with 85% of the energy consumption over the network, compared with existing models. © 2019, Springer Science+Business Media, LLC, part of Springer Nature.