Young-Elderly Gait Classification via PCA Feature Extraction and SVMs

Bjoern M. Eskofier, Peter Federolf, Benno Nigg, and Patrick Kugler


Biomechanical Data Classification, Pattern Recognition, PCA, SVM


The classification of gait patterns has great potential as a diagnostic tool, for example to identify at risk gait in the elderly. This paper presents a method for classifying young-elderly gait via principal component analysis (PCA) feature extraction and support vector machine (SVM) classification. For this purpose, 3D marker trajectories were collected from 36 female subjects walking on a treadmill. PCA dimensionality reduction was directly performed on these trajectories. Using SVMs with linear kernel, a classification rate of 91.7% was achieved. In contrast to other published gait classification methods, this approach does not require prior knowledge of specific time points in the gait cycle (e.g. heel-strike and toe-off) and it does not involve biomechanical models which are usually based on additional assumptions (e.g. joint center positions). Moreover, SVMs with linear kernel allow visualizing the group differences by projecting the normal vector of the decision boundary back onto the original marker space.

Important Links:

Go Back