Due to the few data samples available for computer-aided systems, arrhythmia, a common cardiovascular ailment, creates hurdles for accurate identification. This paper presents a unique methodology for feature extraction in arrhythmia classification that uses a hybrid time-frequency domain analysis within a Convolutional Neural Network (CNN) architecture. The suggested method combines temporal information from RR intervals, frequency-domain information from the Hilbert-Huang transformation, and combined time-frequency information from the continuous wavelet transformation. Following that, a CNN is trained with Focal Loss as the designated loss function. The approach has been carefully tested and verified on an arrhythmia database, proving its ability to classify four unique types of electrocardiogram (ECG) data. Empirical results show that the proposed hybrid time-frequency domain feature extraction method outperforms current classification methodologies in terms of accuracy.