Deep Rule Visualizer: A Novel Explainable Deep Neural Network Visualization Architecture

This paper introduces the Deep Rule Visualizer, an architecture designed to enhance the interpretability of deep neural networks (DNNs). By integrating xDNN model, saliency maps, and IF … THEN rules, the architecture provides users with insights into DNN decision-making. We evaluate the visualizer on benchmark datasets, showcasing its effectiveness in explaining predictions. The user interface displays rule images, regions of interest, and a novel Deep Rule Score (DRS), enabling users to assess prediction reliability. Quantitative evaluation confirms that higher Deep Rule Scores align with correct predictions. The visualizer enhances trust in DNN models, aids error identification, and holds potential across diverse domains.