Bearings are considered as a critical component in Induction Motors (IM). An approach based on Motor current Signature analysis (MCSA) is presented to detect bearing faults (BF). This study is subject for a novel pattern recognition approach for BF detection in IM combining Stationary Wavelet Packet Transform (SWPT) and DAG SVM. Four bearing conditions are tested. As results, it is shown that the proposed approach permits to distinguish with full accuracy different bearing conditions regardless the load level.