In the new 5G and beyond (B5G) connectivity era, Mission Critical Services (MCS) are expected to leverage secure and reliable communication to the end-users. However, diverse network conditions, along with emergency collision events (e.g., abrupt depletion of network or service resources) necessitate a flexible deployment of the MCS, coupled with an efficient management of the associated resources. This work presents a technical solution of a proactive MCS overload detection architecture and methodology, based on the intelligence loop between the MCS and typical 5G core network components. In this context, the monitoring metrics provided by the MCS server are used by the telemetry module for real-time inference using the potency of a pre-trained Machine Learning (ML) model, targeting at forecasting service requirements and providing overload alarms. The automated scalability functionalities of the proposed solution are demonstrated considering a resource overload prediction scenario, so as to intelligently provide notifications about the upcoming needs for resource scaling. To ensure continuous MCS availability in the presence of collision events, the corrective actions by the network Orchestrator include the MCS service scaling by deploying additional pods and providing load balancing capabilities. The regulation of the Deep Neural Network (DNN) hyperparameters and performance comparison against baseline schemes are quantitatively outlined. Conclusively, results provided evidence related to the ML-drivel intelligence loop embracing a successful monitoring of MCS, thereby boosting the reliability and self-configuration in critical conditions.