Quality of Service (QoS) is a general performance measure of user-related services that maintain the reliability and latency without compromising the service quality. The reliable data distribution requires optimization at network layer by processing the massive IoT data. Therefore, Deep Learning (DL) supports massive data process to identify patterns and make decisions. However, concerns related to degree of privacy with QoS guarantees are still unaddressed in Narrowband Internet of Things (NB-IoT). In this paper, the privacy-based hybrid deep learning (PHDL) optimization is carried out on the switching system to improve the network QoS and privacy concern. The DL model enables the optimization of various network parameters including latency, packet delivery rate, and outage probability of NB-IoT in a cell over 10 different users. The DL is modelled in such a way that it reduces the loss rate with improved outage probability and reduced end-to-end latency. The simulation is conducted to test the efficacy of the PHDL with varying degrees of user privacy. The results of simulation shows that the confidential privacy offers improved network and user privacy QoS than other state-of-art models in NB-IoT.