The application of IMUs and artificial neural networks have shown their potential in estimating joint moments in various motion tasks. In this study, IMU data collected with five sensors during gait was used as input data to estimate hip, knee and ankle joint moments using artificial neural networks. Additionally, the original 30 features of the sensors’ data were reduced to their ten most relevant principal components and also used as input to the neural networks to evaluate the influence of feature selection. The prediction accuracy of the networks was lower for the reduced dataset. Research with a larger dataset needs to be undertaken to further understand the influence of a reduced number of features on the prediction accuracy.
Mundt, Marion; Koeppe, Arnd; Bamer, Franz; Potthast, Wolfgang; and Markert, Bernd
"FEATURE SELECTION FOR THE APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN MOTION ANALYSIS,"
ISBS Proceedings Archive: Vol. 38:
1, Article 94.
Available at: https://commons.nmu.edu/isbs/vol38/iss1/94