Vehicle-to-Everything (V2X) communication standards, which mainly rely on the 5G New Radio (NR) technology, can be subject to attacks such as Distributed Denial of Service (DDoS), which flood the network with non-expected control information. This causes network performance degradation and leads to accidents involving vehicles and/or vulnerable road users. A potential approach to mitigate DDoS attacks is to isolate the hijacked vehicular users in sinkhole-type slices that contain a small amount of network resources. Nevertheless, DDoS attacks may be unpredictable since it can modify its communication protocol for example, which makes it difficult to determine the proper moment to release mitigated users from the sinkhole-type slices once the security breach ceases to exist. In such a context, we propose a Reinforcement Learning-based approach that evaluates multiple types of DDoS attacks on sinkhole-type slices and estimates the optimal time to keep a mitigated user in such a slice before releasing it. The proposed approach is trained and tested with a dataset collected from a SG-V2X testbed. Results show that our approach outperforms a benchmark of random actions, in terms of the mean cumulative reward and error over time.