Sensitivity to blockage challenges performance of millimeter-wave (mmWave) communication systems. We apply the MiniRocket machine learning (ML) method to provide reliable early warning of mobile mmWave signal blockage hundreds of milliseconds ahead, thus facilitates a proactive response. MmWave signal datasets for training and testing the ML method are created using our low-complexity physics-based simulation tool, which models diffraction accurately. Our insights and numerical results illustrate that the proposed early warning method is facilitated by the diffraction-induced pre-blockage signal patterns and is robust to diverse environmental and mobility conditions.