Using transfer learning for concurrent multi-alarm prediction in optical transport networks, we assess prediction precision, data volume, and model training time. The method can effectively improve the efficiency of prediction while maintaining accuracy.