Inertial Measurement Units and Partial Least Square Regression to Predict Perceived Exertion During Repetitive Fatiguing Piano Tasks

Aim. Predict the rate of perceived exertion (RPE) of pianists using inertial measurement units (IMUs)-based kinematic descriptors. Method. Fifty expert pianists played
Digital
(right-hand 16-tone sequence) and
Chord
(right-hand chord sequence) excerpts in a continuous loop for 12 min or until exhaustion. Partial least square regression was used to predict RPE with IMUs-based kinematic descriptors. The mean error of prediction over 50 iterations with five-folds cross-validation was used to assess the quality of the model. Variable importance in projection was calculated to determine the most relevant features for predicting the RPE and reduce the number of input predictors. Results. Thirty and twenty-six participants showed signs of fatigue before 12 min of the
Digital
and
Chord
tasks, respectively, and were included in the analysis. The reduced model of 275 and 227 input variables including four-latent variables explained 86.95 ± 0.46 and 83.91 ± 0.54 of the variance of the RPE on the training set with an absolute error of 0.976 ± 0.033 and 1.189 ± 0.068 on the testing set for both
Digital
and
Chord
tasks, respectively. The best features, variables, and sensor positions to predict RPE were different between both
Digital
and
Chord
tasks suggesting a task-dependency in the prediction of effort exertion during piano performance. Conclusion. These results highlight the feasibility of continuously monitoring RPE in pianists using kinematic descriptors. These results are promising for developing methods to prevent high levels of fatigue and injuries in musicians.