Explainable Deep Learning for Lung Disease Detection on Chest X-Ray Images Using Shapley Additive Explanations (SHAP)

COVID-19, Pneumonia,Citra X-ray, Explainable Artificial Intelligence, Shapley Additive Explanations

Authors

  • Muhammad Irsyad Informatics Engineering Study Program, Faculty of Science and Technology, Universitas Islam Negeri Sultan Syarif Kasim Riau, Pekanbaru-Riau, Indonesia.
  • Benny Sukma Negara Informatics Engineering Study Program, Faculty of Science and Technology, Universitas Islam Negeri Sultan Syarif Kasim Riau, Pekanbaru-Riau, Indonesia.
  • Sarifah Muliani Informatics Engineering Study Program, Faculty of Science and Technology, Universitas Islam Negeri Sultan Syarif Kasim Riau, Pekanbaru-Riau, Indonesia.
February 9, 2026
February 12, 2026

Downloads

X-ray imaging is an effective method for detecting lung diseases, such as COVID-19 and pneumonia. With the advancement of technology, the diagnosis process can now be carried out more accurately by utilizing artificial intelligence-based systems. One of the most widely used methods is deep learning, but this method is black-box, making it difficult to understand the reasons behind the model's decisions. The purpose of this study is to build an X-ray image classification system using a deep learning model based on a Convolutional Neural Network (CNN) with a VGG-16 architecture, and to apply the Shapley Additive Explanations (SHAP) method to provide explanations regarding the visuals related to the image areas that affect the prediction results. The model was trained using several configurations, and the best results were obtained with a data ratio of 80%:20%, a learning rate of 0.001, a batch size of 32, and 50 epochs. The results showed that the model was able to achieve an accuracy of 95.75% on the training data and 96.00% on the validation data. The SHAP method was used to improve understanding of the model's predictions and can display feature contributions in the form of a heatmap. This makes it easier to understand which areas have the most influence on the prediction results. The results show that the combination of deep learning and SHAP is capable of providing visual explanations for the model's prediction results.