This paper proposes the load prediction technique of cluster in AI-based ship combat system. Within the cluster, live migration is performed by transferring a running virtual machine to another host without service interruption. At this point, the CPU usage ranking is calculated and the migration is based on the percentage of the same rank. However, simply migrating to the rank of CPU usage results in unnecessary migrations within the cluster, even though enough idle resources inside the CPU are available, if repeated until the CPU resource utilization ranking is as equitable as possible. To solve these problems, this paper designs the overload prediction model of AI-based clusters, thereby improving live migration efficiency and stability.