Bearing fault diagnosis is crucial for maintaining the reliability and efficiency of rotating machinery in industrial settings. This research develops an innovative approach to intelligent bearing fault diagnosis by combining Feature Mode Decomposition (FMD) with advanced machine learning techniques. We address the challenges of analyzing non-stationary signals and detecting multiple fault conditions simultaneously. The proposed method introduces a novel mode selection criterion, the Envelope Kurtosis-Entropy Product (EKEP), to identify the most informative mode from FMD outputs. Comprehensive feature extraction is performed on the selected mode, followed by applying various machine learning algorithms for fault classification. Experimental validation using a Spectra Quest Machinery Fault Simulator demonstrates good classification performance across different machine learning models, achieving superior results in diagnosing multiple bearing fault conditions, including combined faults.