This document describes research on using a triaxial accelerometer located on a wrist-worn smartwatch to detect falls in elderly people. Researchers collected acceleration data from simulated falls (both syncope and forward falls) and other activities to train and evaluate several machine learning classifiers. Their best-performing models achieved over 95% accuracy in distinguishing fall from non-fall events with as few as 7 attributes. Evaluation on data from new subjects also showed robustness, with over 90% accuracy. The researchers conclude the approach is promising but note limitations from battery life and plan to explore using additional sensors for fall detection.