Towards Explainable Artificial Intelligence for Medical Text Comprehension

Abstract

This work focuses on the automatic classification of cancer-related biomedical documents uti lizing both classical Machine Learning and Deep Learning approaches. The study leverages the Biomedical Text Publication Classification dataset from Kaggle, with the goal of identifying the most effective methods for accurately and reliably classifying oncology texts. The proposed approach evaluates several models, such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Transformer-based architectures, and traditional machine learning me thods based on Bag-of-Words (BoW) and TF-IDF representations. Experimental results indicate that the XGBoost classifier combined with TF- IDF features achieves the best overall performance among the evaluated models. Performance evaluation is conducted using multiple metrics, including accuracy, precision, re call, F1-score, and the Receiver Operating Characteristic (ROC) curve. Additionally, the SHAP (SHapley Additive exPlanations) method is employed to enhance model interpretability by iden tifying the most influential features contributing to predictions. Finally, a web-based decision support system is developed to assist healthcare professionals in the automatic classification of medical reports and to provide statistical visualizations across three cancer types : lung, thyroid, and colon cancer. This work underscores the importance of integrating effective feature representation, robust machine learning models, and interpretability techniques in medical text classification tasks.