Base station deployment costs pose a significant challenge for operators, especially in regions without 5G infrastructure. Sharing radio access networks (RANs) has emerged as a promising solution since it enables operators to lower installation costs by sharing redundant and available resources. Open-RAN (O-RAN) is a new RAN framework that aims for intelligence and openness in hardware and software RAN sharing with the help of virtualization technology and disaggregated architecture of RANs. Multiple operators could coexist together and their virtualized RAN components can be deployed on each other's computing resources. In this disaggregated architecture, MAC scheduler fundamentally governs resource allocation to users associated with a base station and resides in RAN's distributed unit (DU) that can be virtualized and deployed on O-RAN. Traditionally, MAC scheduling is handled by static methods that makes its adaptation to dynamic environments challenging. While Deep Reinforcement Learning (DRL) offers a promising solution to MAC scheduling; but, a global network view is necessary for adapting new traffic patterns. However, information sharing between operators compromise privacy and competition between operators. Therefore, in this study, we explore the use of Federated learning-based DRL (FDRL) for MAC scheduling in RAN sharing in O-RAN.