Dual connectivity is essential for sixth-generation (6G) and future networks, with carrier aggregation (CA) serving as the key technology that enhances data rates by dynamically allocating multiple frequency carriers to users. However, the constant change of carriers can introduce excessive handovers, leading to signaling overhead and latency issues. In this work, we propose a two-step carrier switching mechanism where the first step predicts the throughput and the number of handovers for given carrier switching decision parameters using different artificial intelligence (AI) mechanisms. The second step is to choose the policy that provides the best trade-off between the number of handovers and the median UE throughput. We utilize real-world traffic data on an industry-grade network simulator to train and test the proposed carrier switching algorithm. We compare four AI models in the first step: random forest, transformer, long-short-term memory (LSTM), and fully connected neural network. Our analysis demonstrates that random forest achieves superior prediction performance for median throughput and average handover frequency due to its robustness against dataset variations. Our results indicate that the number of handovers can be reduced by 40% with a 3% reduction in throughput compared to the handover-unaware baseline.