Cross-Validation results for a gait phase estima-tion with Artificial Neural Networks

Abstract

The knowledge of the gait phase can improve the support of an active prosthesis to increase the mobility of people with transtibial amputations. A Cross-Validation is used in this work to evaluate a novel approach for continuous gait phase estimation with Artificial Neural Networks. The estimation of the gait phase only uses kinematic variables of the shank, which are measurable by a single Inertial Measurement Unit placed at the shank. The dataset is separated in training data, validation data and test data with a Leave-P-Groups-Out-Approach. With the results, a statement can be made whether the dataset is big enough, or needs additional subjects. With the exemption for one test subject the choice of test subject does not change the quality of the regression significantly.