A robust XGBoost approach for bearing fault prediction from vibration signals: Feature analysis and model evaluation