The electrocardiogram (ECG) signal feature extraction and classification diagnosis algorithm is proposed to address the high incidence of heart disease and difficulty in self-detection. First, the collected ECG signals are preprocessed to remove the noise of the ECG signals. Next, wavelet packet decomposition is used to perform a four-layer transformation on the denoised ECG signal and the 16 obtained wavelet packet coefficients analyzed statistically. Next, the slope threshold method is used to extract the R-peak of the denoised ECG signal. The RR interval can be calculated according to the extracted R peak. The extracted statistical features and time domain RR interval features are combined into a multi-domain feature space. Finally, the particle swarm optimization algorithm (PSO), genetic algorithm (GA), and grid search (GS) algorithms are applied to optimize the support vector machine (SVM). The optimized SVM is utilized to classify the extracted multi-domain features. Classification results show the proposed algorithm can classify six types of ECG beats accurately. The classification efficiency achieved by PSO, GA, and GS are 97.78%, 98.33%, and 98.89%, respectively.