In the present study, entropy values of EMG series, collected from arms and legs of healthy volunteers by using 8 recording channels in both normal and agressive actions, by using six different methods (Lempel-Ziv Entropy, Shannon Entropy (ShanEn), Logarithmic Energy Entropy, Approximate Entropy, Sample Entropy, Permutation Entropy (PermEn)) have been classified with respect to three feature sets (all channels, only arms, only legs). In classification step; nonlinear Support Vector Machines) with 5-fold cross validaiton was examined. The results show that stimulus parameters that stimulate muscle cells affect the level of complexity of EMG series. ShanEn and PermEn provide the best performance in classifying physical actions by means of EMG. The performance of both approaches have been improved by using Ensemble Learning with marginal function in classifying contrast physical actions. Measurements must be segmented to analyze EMG series.