The proposed research work demonstrates a systematic procedure for studying the chaotic dynamics of normal and aggressive physical actions by analysing the EMG (Electromyogram) signals in the fractal realm. The intrinsic patterns of the EMG signals are extracted through the fractal features of the signal. The fractal features such as Higuchi Fractal Dimension (HFD), Maximum Fractal Length (MFL) and Lacunarity are computed from the Surface EMG (sEMG) signal. These features derived by fractal analysis are adopted for signal characterization, which is successfully administered to classify sEMG signals. The proposed feature extraction and classification techniques have been validated using sEMG signals obtained from the UCI Machine Learning Repository. Based on the inferences made, it can be validated that fractal features provide an optimal diagnostic feature set to classify sEMG signals into various action classes.