High rate Global Navigation Satellite System (GNSS) deformation time series capture a broad spectrum of earthquake strong motion signals for rapid contributions to hazard warnings and assessment, but experience regular sporadic noise that can be difficult to distinguish from true seismic signals. Previous studies developed methods for automatically detecting these signals but most rely on various external inputs to differentiate true signal from noise. In this study we generated a dataset of high rate GNSS time differenced carrier phase (TDCP) velocity time series concurrent in space and time with expected seismic surface waves from known seismic events. TDCP velocity processing has increased sensitivity relative to traditional geodetic displacement processing without requiring sophisticated corrections. We trained, validated and tested a random forest machine learning classifier. We find our supervised random forest classifier outperforms the existing detection methods in stand-alone mode by combining frequency and time domain features into decision criteria. We optimized the classifier on a balance of sensitivity and false alerting. Within a 100km epicentral radius, the classifier automatically detects 86% of events greater than MW5.0 and 98% of events greater than MW6.0. The classifier model has typical detection latencies seconds behind S-wave arrivals when run in real-time mode on “unseen” events. We conclude the performance of this model provides sufficient confidence to enable these valuable ground motion measurements to run in stand-alone mode for development of edge processing, geodetic infrastructure monitoring and inclusion in operational ground motion observations and models.