In this paper, a random forest regressor (RFR)-based monitoring scheme is proposed to monitor the mutual inductances of inductive power transfer (IPT) systems. RFR is a supervised machine learning (ML) algorithm, which gains the merits of strong robustness against disturbance. The proposed monitoring scheme inherits this characteristic, and therefore exhibits more accurate monitoring performance than conventional model-based monitoring schemes for practical IPT systems with parameter drifts. The RFR of the proposed method is trained based on the primary-side voltage and current at the resonant frequency rather than the primary-side measurements at the sub-resonant frequencies of the conventional model-based methods. Therefore, real-time monitoring of mutual inductances for IPT systems can be achieved, which is validated by experimental results.