A Neuro-Fuzzy Model for Predicting EMG of Trunk Muscles Based on Lifting Task Variables

This paper discusses development of a neuro-fuzzy expert model for predicting electromyographic (EMG) responses of trunk muscles in manual lifting based on task. The model utilizes two task variables, i.e. trunk moment and trunk velocity, as inputs, and ten muscle activities as outputs. The input and output variables are represented using the fuzzy membership functions. Initial fuzzy rules are generated by neural networks using true EMG data. The refined fuzzy rules are used to derive the prediction model. The model was developed based on EMG data for 8 subjects, and validated using the EMG data for another 4 subjects. The model allowed to predict the normalized EMG values with the mean absolute error ranging from 4.97% to 13.16% (average = 8.43%, SD=2.87%), and average value of the mean absolute difference between the real and predicted EMG of 6.4% (SD=3.39%). It is concluded that prediction of EMG responses in manual lifting tasks is feasible, and that model performance could be improved by increasing the number of lifting task variables.