Towards Better Performance in Incomplete LDL: Addressing Data Imbalance