Cardiovascular diseases remain the leading cause of mortality worldwide, placing a substantial burden on healthcare systems. In many cases, early symptoms are subtle or absent, making timely detection challenging. Electrocardiography (ECG) is a widely used, non-invasive diagnostic method for detecting cardiac abnormalities; however, manual interpretation is time-consuming, requires specialized expertise, and is prone to subjectivity. These challenges highlight the urgent need for automated, accurate, and interpretable diagnostic systems. This work offers a heart illness recognition system utilizing ECG signals, integrating contemporary deep learning models with the Fuzzy C-Means (FCM) method to enhance accuracy and interpretability in response to the pressing demand for automation. Three model architectures were specifically created and evaluated: (1) CNN + Transformer, (2) LSTM + Attention, and (3) CoAtNet 1D—a hybrid architecture integrating convolutional and self-attention methods. The models were trained on the PTB-XL ECG dataset to categorize five principal types of cardiovascular conditions: Myocardial Infarction (MI), Ventricular Hypertrophy (HYP), Conduction Disturbance (CD), ST/T Wave Change (STTC), and Normal (NORM). Experimental findings indicate that all three models attained accuracy over 96.5%. The incorporation of the FCM algorithm during post-processing facilitates the identification of anomalous ECG signal segments in a visually led and data-driven approach, aiding physicians in comprehensive analysis and clinical decision-making. The suggested approach illustrates practical viability and underscores the significant promise of data-driven technologies in intelligent healthcare.