The purpose of this study was to examine the use of a wrist-worn triaxial accelerometer-based activity monitor for classifying upper- and lower-body dumbbell RT exercises.Methods
Sixty participants performed 10 repetitions each of 12 different upper- and lower-body dynamic dumbbell exercises. Algorithms for classifying the exercises were developed using two different methods: support vector machine and cosine similarity. Confusion matrices were developed for each method, and intermethod reliabilities were assessed using Cohen’s kappa. A repeated-measures ANOVA was used to compare the predicted repetitions, identified from the largest acceleration peaks, with the actual repetitions.Results
The results indicated that support vector machine and cosine similarity accurately classified the 12 different RT exercises 78% and 85% of the time, respectively. Both methods struggled to correctly differentiate bench press versus shoulder press and squat versus walking lunges. Repetition estimates were not significantly different for 8 of the 12 exercises. For the four exercises that were significantly different, the differences amount to less than 10%.Conclusion
This study demonstrated that RT exercises can be accurately classified using a single activity monitor worn on the wrist.