5G networks comprise a dense network of access points to mitigate the reduced coverage problem resulting from using high-frequency ranges such as mm-waves. Using these frequencies alleviates the problem of bandwidth scarcity as well. However, one of the challenges in this area is for the users to be able to select an efficient access point that benefits them in terms of meeting their QoS requirements, such as delay; and also reduce the random access channel congestion that occurs at the access points. In order to solve this problem, we first establish an optimization problem and experiment with a reinforcement learning-based scheme. We assess the results in terms of random access performance metrics. Experiment results demonstrate the effectiveness of the approach. In particular, implementing a reinforcement learning technique allowed a 44.5% reduction in average access delay and improved access success probability.