This document presents an optimization of algorithms for real-time ECG beat classification. It compares algorithms using voltage values in the time domain versus those using Daubechies wavelet analysis. It extracts features around reference peaks within the QRS complex and uses clustering methods to classify beats in real-time as normal, premature ventricular contraction, or unclassified. Evaluating algorithms on 32 MIT-BIH records, the method using Daubechies wavelets and correlation measure achieved 93.25% sensitivity and 91.43% positive predictivity for premature ventricular contraction detection, making it suitable for real-time systems due to low computational cost.