Neural Network Optimization for Estimating Lower Limb Angular Velocity Based on Body Mass Index and Walking Speed

The angular velocity of lower limb joints is an important kinematic parameter that represents gait dynamics and is widely used as a reference in walking rehabilitation systems as well as assistive robot control. This study aims to optimize the neural network architecture to estimate the angular velocity of the hip, knee, and ankle joints by taking into account the Body Mass Index (BMI) and walking speed. The study was conducted using secondary gait data grouped based on BMI category, joint type, and gait phase. The estimation model was developed using a feedforward neural network with one hidden layer, where the number of neurons was varied from 5 to 30. The results showed that increasing the number of neurons in the hidden layer improved estimation accuracy until reaching an optimal condition in the range of 15–20 neurons, after which performance tended to saturate. The hip joint showed the most stable and accurate estimates, followed by the knee and ankle, while the late gait phase resulted in higher accuracy compared to the early gait phase. These findings emphasize the importance of optimizing neural network architecture to produce accurate and efficient gait parameter estimations for walking rehabilitation applications.

Keywords: Angular Velocity, BMI, Gait, Neural Network, Robot Rehabilitation, Walking Speed