An individualized gait pattern prediction model based on the least absolute shrinkage and selection operator regression

Gait pattern prediction is an essential function of individualized motion control of the lower-limb exoskeleton. This paper presents a novel gait pattern prediction model based on the least absolute shrinkage and selection operator (LASSO) regression. Gait data were collected from one hundred and twenty healthy young adults (78 males and 42 females), who were instructed to walk back and forth on a 12-meter-long walking platform while having their heel strike coinciding with the beat of a metronome. The lower-limb joint (i.e., the hip, knee and ankle) angular kinematics were segmented, resampled and transformed into Fourier coefficients. The LASSO regression model with age, gender and 14 anthropometric parameters as prediction variables was trained and used to estimate the Fourier coefficients which were then applied in the lower-limb joint angle trajectory reconstruction. The results showed that the root mean square errors between the actual and predicted joint angle trajectories ranged from 3.41° to 4.55°. The parameters of the linear fit method further revealed the waveform similarity between the actual and predicted joint angle time series. These results suggested that the proposed model was able to accurately predict lower-limb joint kinematics during gait. Application of the proposed model can help resolve the overfitting problem, and provides a new solution to individualized gait pattern prediction.