Nonlinear component analysis as a kernel eigenvalue problem
Neural Computation
WSEAS Transactions on Signal Processing
| Hi-index | 0.00 |
The quantitative assessment of the change of gait symmetry has played a very important role in the clinical diagnostics. This paper investigated the application of an advanced hybrid machine learning approach such as the combining kernel-based principal component analysis (KPCA) with support vector machine (SVM) to evaluate the change of gait symmetry quantitatively based on the basic idea that the discrimination of the functional change of between human lower extremities can be hypothesized as binary classification task. To assess the change of gait symmetry accurately, more nonlinear principal components extracted by using KPCA were employed to initiate the training set of SVM, which could enhance the generalization performance of SVM. The foot-ground force gait data of 24 elderly participants were acquired using a strain gauge force platform during normal walking, and were analyzed with our proposed model. The test results demonstrated that, when compared to the SVM-based classification models, our proposed technique with superior classification performance could discriminate difference between the right and left side gait function of lower limbs accurately, and that more principal components extracted by KPCA with polynomial kernel (d=3) could capture more useful information about intrinsic nonlinear dynamics of human gait in comparison to the some key gait variables selected. The proposed hybrid model could function as an effective tool for clinical diagnostics in the future clinical circumstance.