Testing of integrated circuits (IC) is a highly expensive process but also the most important one in determining the defect level of an IC. Manufacturing defects in the IC are modeled using stuck-at-fault models. Stuck-at-fault models cover most of the physical faults that occur during the manufacturing process. With decreasing feature sizes due to the advancement of semiconductor technology, the defects are also getting smaller in size. Tests for these hard-to-detect defects are generated using deterministic test generation (DTG) algorithms. Our work aims at reducing the cost of Path Oriented Decision Making: PODEM (a DTG algorithm) without compromising the test quality. We trained a meta predictor to choose the best model given the circuit and the target net. This ensemble chooses the best probability prediction model with a 95% accuracy. This leads to a reduced number of backtracking decisions and much better performance of PODEM in terms of its CPU time. We show that our ML- guided PODEM algorithm with a meta predictor outperforms the baseline PODEM by 34% and other state-of-the-art ML-guided algorithms by at least 15% for ISCAS85 benchmark circuits.