This paper investigates network slicing in radio access networks (RANs) that support multiple virtual operators, focusing on maximizing operator utility and ensuring service isolation on a shared physical platform. Network slicing enables efficient resource management and allows for flexible scaling of services based on diverse operator requirements. We evaluate RAN resource slicing for virtual operators by addressing two key challenges: resource reservation and allocation. Operators predict future resource demands using artificial neural networks and reserve a portion of physical resources accordingly. Following reservation, the allocation process is modeled as a Markov decision process aimed at maximizing revenue, which is solved using reinforcement learning algorithms. This dual optimization approach ensures utility maximization and efficient service delivery for virtual operators while balancing network load and enhancing system robustness in dynamic environments. Additionally, our approach accounts for fluctuating user traffic patterns and varying operator priorities, demonstrating significant potential for improving resource utilization, maintaining service-level agreements, and fostering scalable, resilient network operations in highly virtualized environments. Through comprehensive simulation, we establish that the proposed method leads to a significant improvement in resource utilization and operator utility across various traffic loads and operator priority scenarios, clearly demonstrating its competitive advantage in the context of resource reservation and allocation for network slicing in virtualized environments.