C4.5 Decision Tree against Neural Network on Gait Phase Recognition for Lower Limp Exoskeleton

Feedbak control of exoskeleton suit requires recognition of gait phases occurred in one gait cycle. This paper presents C4.5 decision tree algorithm compared with multilayer perceptron (MLP) and nonlinear auto regressive with exogenous variable (NARX) recognizing three different gait phases of a passive lower limp exoskeleton: Stance, Swing, and Push. Two IMU sensors on hip and knee joint and two FSR sensors in a self-made shoe provide four inputs for the algorithms. Test data at different walking speeds are collected. The experimental results show classification success rate of each algorithm when algorithm trained with different size of pattern. The result of implementation on Arduino is verified.