Optimization of Sensor Number for Lower Limb Prosthetics using Genetic Algorithm

Research in assistive robotics has been in constant development for the last two decades. One of the main topics in this field is gait prediction for lower limb prosthetics. This paper aims to explore the effect of using genetic algorithm with machine learning for reducing the number of sensors and/or input signals for determining gait activities. This technique is applied on a sample of a published dataset of lower limb neuromechanical signals due to its wide range of used sensors and sensor signals. This research focuses on sensor optimization, i.e., maintaining same accuracy with fewer number of sensors, rather than optimizing the machine learning model itself or facing model problems such as overfitting. Error measurements of different machine learning methods are compared before and after using the genetic algorithm. Results showed that, for K-Nearest Neighbor, Random Forest, Decision Tree and Gaussian Naïve Bayes, sensors/input signals can be reduced without a significant effect on the accuracy.