An inefficient utilization of network resources in a time-varying traffic environment often leads to load imbalances, high call-blocking events and degraded quality of service (QoS). This paper optimizes the QoS of a cloud radio access network (C-RAN) by investigating load balancing solutions. The dynamic re-mapping ability of C-RAN is exploited to configure the remote radio heads (RRHs) to proper base band unit sectors in a time-varying traffic environment. RRH-sector configuration redistributes the network capacity over a given geographical area. A self-optimized cloud radio access network (SOCRAN) is considered to enhance the network QoS by traffic load balancing with minimum possible handovers in the network. QoS is formulated as an optimization problem by defining it as a weighted combination of new key performance indicators for the number of blocked users and handovers in the network subject to RRH sectorization constraint. A genetic algorithm (GA) and discrete particle swarm optimization (DPSO) are proposed as evolutionary algorithms to solve the optimization problem. Computational results based on three benchmark problems demonstrate that GA and DPSO deliver optimum performance for small networks, whereas close-optimum is delivered for large networks. The results of both GA and DPSO are compared to exhaustive search and $K$ -mean clustering algorithms. The percentage of blocked users in a medium sized network scenario is reduced from 10.523% to 0.421% and 0.409% by GA and DPSO, respectively. Also in a vast network scenario, the blocked users are reduced from 5.394% to 0.611% and 0.56% by GA and DPSO, respectively. The DPSO outperforms GA regarding execution, convergence, complexity, and achieving higher levels of QoS with fewer iterations to minimize both handovers and blocked users. Furthermore, a tradeoff between two critical parameters for the SOCRAN algorithm is presented, to achieve performance benefits based on the type of hardware utilized for C-RAN.