The effect of network depth in neural network for human gait cycle prediction

Artificial neural networks were implemented satisfactorily to assess gait events from various walking data. This research is to study the suitable network depth in neural network technique for developing human gait cycle prediction model using artificial neural network. Gait dataset is retrieved from public dataset where it is measured from 24 young adults who in the last six months before the data was collected and had no lower-extremity injury and were all free of any orthopedic or neurological diseases that could interfere with their gait patterns. In Artificial Neural Network (ANN) developed model, the depth of neural network is one of factor that determine the performance of the developed model. The performance of the model will be compared in terms of Regression (R) and Mean Square Error (MSE) value. To develop human gait prediction model, the input variable is joint angle and joint moment for hip, ankle, and knee. Moreover, only sagittal plane which is Z-axis is used in this study. A multi-layer perceptron model is implemented, composed with different hidden layers and hidden neurons. With 10th hidden layers attempt, on the 8th hidden layers, the R-Value of gait cycle prediction was 94% for training 95% for testing. And the lowest testing Root Means Square Error (RMSE) is at 59.87. The role of ANN in the prediction gait cycle is discussed in this paper.