Predictions of seat-to-head transmissibility based on Bayesian-optimized neural network and feature selection

Low-frequency vibrations transmitted to the head could cause the riding discomfort and the motion sickness. Accurately predicting the seat-to-head transmissibility was essential for enhancing the riding comfort. The radial basis function neural network model to predict the seat to head transmissibility was developed using the Bayesian optimization based on the whole-body vibration experiments. The L1-norm and the mutual information algorithms were applied to select the subject physical characteristics that influence the seat-to-head transmissibility, thereby addressing the coupling effects among subject physical characteristics and improving the prediction accuracy of the model. It was found that the predictive models based on the radial basis function neural network with the feature selection optimization conducted through the L1-norm and the mutual information algorithms selected the gender, the age, and the body mass index as the important subject physical characteristics in the human vibration model, thereby enhancing both the computational efficiency and the predictive accuracy of the seat to head transmissibility. The model with the mutual information feature selection achieved higher prediction accuracy for the seat-to-head transmissibility than the L1-norm-based mode.