Figure 1.
Neural network architecture representing the three layers as well as the tangential sigmoid and pure linear transfer functions in the hidden and output layers, respectively.
All nodes are not represented in this diagram, though a weighted sum of all inputs and the bias is performed at each node in the hidden and output layers.
Figure 2.
Number of epochs required for convergence to error goal given the number of hidden nodes during training of the neural network with all 16 input variables.
Table 1.
Demographics of all 56 participants [mean (SD)].
Figure 3.
Performance of the neural network using all 16 input variables.
Figure 4.
Maximum mapping performance of a three layer neural network in estimating the CoM-BoS distance, CoMv-Bos displacement and BoS area across the five different input variable categories as well as when using a combination of all input categories.
Figure 5.
Representative data for the CoMv-BoS distance (A) and the BoS Area (B), as calculated by a neural network (triangles) with 20 hidden nodes and an error goal of 0.01.
All input variables were included in this training set, with the actual values for these balance control measures represented by the open circles.
Table 2.
Average performance (SD) of selected combinations of inputs and the corresponding hidden nodes and error goal values that produced the highest accuracy.