The wavelet packet based filtering/denoising performance is analyzed by using Balance Sparsity-norm & fixed form thresholding (soft &hard) methods where the Mean, Standard Deviation (SD) & Mean Absolute Deviation (MAD) is calculated at different global threshold for healthy, myopathic & neuropathic EMG signals. The intension is to extract the residuals of healthy and diseased EMG signals which provide the significant results for classification of healthy, myopathic & neuropathic EMG signals. The features are extracted or the coefficients are generated using “haar-3”. These two methods have a fairly large accuracy percentage which can be used as a diagnostic tool in medical field. The technique mentioned in this paper is a mathematical tool for the detection of myopathy and neuropathy as compared to the conventional instrumental ones. Hence, it is faster, efficient and robust as it is resistant to environmental hazards.