Explainable Deep Learning Model for Cardiac Arrhythmia Classification

In this work, we proposed a hybrid deep learning model that (CNN-GRU) combines a One-Dimensional Neural Network (1D CNN) and a Gated Recurrent Unit (GRU) to classify four types of cardiac arrhythmia and applied LIME to provide explanations for its predictions. LIME is a well-known local explanation method that can explain any machine learning model by simulating its behaviours to generate explanations. However, LIME can only explain tabular, text, and image datasets. Therefore, we proposed a visual presentation of LIME on signal dataset by applying a heatmap to highlight important areas on the heartbeat signals. Moreover, we propose an effective method to segment heartbeats from ECG records, ensuring that all key features are extracted correctly, such as QRS Complex, P Wave, and T Wave. The proposed hybrid model was trained using ECG lead II from the MIT-BIH dataset and evaluated based on accuracy, precision, recall, f1 score, and AUC-ROC performance matrix. To highlight the proposed model's validity, we compare it against the standalone CNN and GRU models and prove its superiority in terms of accuracy and ROC.