2.11. Model evaluation

The evaluation of our modeling approach was based on the evaluation practices suggested for OpenSim models. The assessment of the quality of the model's inverse kinematics, for each of the subjects for each session, was based on the root mean square (RMS) function that is built in Opensim that calculates the RMS between the experimental marker positions and virtual positions extrapolated from the calculated kinematics. For FW the average RMS was kept under 1 cm and the maximum marker error was less than 2 cm, for BW these values were 2 and 4 cm respectively, which complies with OpenSim guidelines. To evaluate quality of the joint moments prediction performed by CEINMS, the R² measure between the joint moments obtained by applying the inverse dynamics to the experimental data and the joint moments predicted by CEINMS was calculated. We found values of R² exceeding 0.9 on average across subjects and experimental sub-phases.