Skip to main content

Table 1 Comparison of linear regression and neural network energy expenditure estimates made per gait cycle for different use cases during assisted walking

From: Rapid energy expenditure estimation for ankle assisted and inclined loaded walking

Model Metric Novel Conditiona Novel Subjectb Both-Novelc
Linear Regression RMSEd\(\left (\frac {\mathrm {W}}{\text {kg}}\right)\) 0.18 0.43 0.41
  Error 4.1% 8.4% 8.2%
Neural Network RMSE \(\left (\frac {\mathrm {W}}{\text {kg}}\right)\) 0.24 0.40 0.43
  Error 4.4% 8.0% 8.1%
  1. aThe novel condition use case randomly selected 10% of the conditions from any subjects as a test set, this was repeated as many times as there were subjects, with performance averaged across test sets
  2. bThe novel subject use case removed one subject at a time from the training set to be the test set, averaging the performance across all subjects
  3. cThe both-novel use case removed one subject at a time as well as two random conditions across all subjects from the training set. These removed conditions were estimated for the test set subject, with results averaged across all test sets
  4. dRMSE is the root mean squared error normalized by the average subject mass