In addition to requiring the accuracy and computational efficiency for solving low-frequency subsurface sensing problem on the airborne transient electromagnetics (ATEMs), to the best of our knowledge, the complexity of subsurface sensing problems should also be considered in order to reduce more and more computational resources, particularly for a large-scale complicated multis-cale problem with a difference between background and targets. For simulating the open-domain, the finite-thickness perfectly matched layer is used to truncate the computational region, while the whole domain becomes larger so that the problem turns more complex. As a result, we propose a novel perfectly matched monolayer (PMM) model based on the extreme gradient boosting (XGB), which is selected and added to further improve the performance during the finite-difference time-domain (FDTD) simulation. The proposed XGB-based PMM model can achieve higher accuracy by using the ensemble learning method of feature attention, and has less memory and time consumption at the same time. Besides, this model has significant advantages in terms of model training stability and its lightweight due to the fact that it relies on the characteristics of traditional machine learning models. Finally, three-dimensional numerical simulations of ATEM problems are carried out to prove the validity and stability of the proposal. The proposed model can not only achieve advantages in numerical accuracy, efficiency and problem complexity, but also be integrated into the FDTD solver to deal with the low-frequency ATEM problems.