In cellular-D2D networks, users can select the communication mode either direct and form D2D links or indirect and communicate with BS. In former case, users should perform pairing selection and choose their pairs. The main focus in this paper is proposing an analytical framework by using tools from stochastic geometry to address these two issues, i.e. i) mode selection for the user devices to be established in either cellular or D2D mode, which is done based on received power from BS influenced by a bias factor, and ii) investigation of choosing n th -nearest neighbor as the serving node for the receiver of interest, by considering full-duplex (FD) radios as well as halfduplex (HD) in the D2D links. The analytic and simulation results demonstrate that even though the bias factor determines the throughput of each mode, it does not have any influence on the system sum throughput. Furthermore, we demonstrate that despite of suffering from self-interference, FD-D2D results in higher system sum throughput as well as higher coverage probability in comparison to its counterpart, namely purely HD-D2D network.