I used a random forest to classify actions using wearable sensor data from various individuals in the Weight Lifting Exercise Dataset (also known as the Human Activity Recognition / HAR dataset). As literature and Kaggle competitions have shown, this results in a highly accurate classification, with our out-of-bag (OOB) and cross-validation error rate suggesting that our test set error rate will be <1% (which is confirmed by an actual run against our test set where we see 20/20 classifications correct).
This is for the John Hopkins Bloomberg School of Public Health Coursera Data Science Specialization.