Electrocardiogram-based biometric identification is still relatively new. In this paper, we have realized an identification system composed of three main phases: data preprocessing, feature extraction, and classification. First, we proposed to use an R-centered segmentation method. Then, the fusion of different types of features is proposed in this work, including zero-crossing rate, entropy, and cepstral coefficients. Their fusion improved the individual identification rate in this system. For classification, we proposed to apply classical learning methods, such as K-Nearest Neighbors (KNN) and MultiLayer Perceptron (MLP), on the extracted features. Finally, we proposed a combination of classical learning methods and deep learning methods, such as CNN+KNN, to identify individuals. This paper proposes a person identification system considering both normal and abnormal ECG signals using both sets of data, such as the PTB-D (healthy individuals) database and the MITBH-A (individuals with arrhythmic signals) database. The experimental results outperform other models in terms of accuracy, demonstrating high identification performance (98.46%) on the PTB-D database (normal heartbeats) and (95.74%) on the MITBH-A database (arrhythmic heartbeats).