In this paper, we consider a multi-user downlink non-orthogonal multiple access (NOMA) network comprising of an access point (AP), multiple near-user (NU) nodes and multiple far-user (FU) nodes. The NUs and FUs exist in clusters of size $N$ and F, respectively. Using limited channel state information (CSI) knowledge, intelligent NU and FU selection coupled with mode switching between NOMA and orthogonal multiple access (OMA) is applied to improve the system throughput and energy efficiency. Considering the case of fixed power allocation (FPA) and dynamic power allocation (DPA), closed-form throughput expressions are derived for the first time in literature with mode selection and user selection. It is seen that power allocation with very limited CSI provides huge gains in throughput and saves power at the AP, and thereby increases the energy efficiency. We validate the correctness of analytical results and the superiority of the considered scheme through Monte-Carlo simulations.