Explainable AI in Healthcare: Text Classification Using Bidirectional LSTM and LIME

Abstract

Although deep learning models can help to improve clinical text classification significantly, their poor interpretability made them unsuitable for adoption in the medical settings. Unlike the past studies, this research adopted the explainable artificial intelligence (XAI) technique to improve the interpretability of the bidirectional long short-term memory (Bi-LSTM) model for clinical text classification. In particular, the study selected the local interpretable model-agnostic explanations (LIME) technique as the XAI technique to interpret the Bi-LSTM model used to classify 24 classes of diseases. The proposed Bi-LSTM model was developed based on the following three main stages: (1) preprocessing; (2) classification; and (3) model explanation. Data for symptoms and diseases were gathered from the Kaggle and used to train and evaluate the model. The outcomes of the study showed that the model achieved precision, recall, F1-score, and accuracy of 98.00%. In addition, the model was further reevaluated by using two additional datasets. By utilizing the 40-class medical transcription dataset, the model attained 76.18% accuracy with F1-score of 0.82. However, by adopting the 30-class symptom-based disease dataset, it accomplished 95.26% accuracy with F1-score of 0.95. The results further unveiled that the LIME technique was not only able to help understand the decision-making process of the model by providing the local explanations, but it also acted as an active diagnostic instrument for refining the preprocessing pipeline during the development of the model.

Received: 19 April 2026 | Revised: 2 July 2026 | Accepted: 14 September 2026

Conflicts of Interest

The authors declare that they have no conflicts of interest to this work.

Data Availability Statement

The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/niyarrbarman/symptom2disease, https://www.kaggle.com/datasets/tboyle10/medicaltranscriptions, and https://www.kaggle.com/datasets/shoaibuddin/symptoms-based-medical-test-recommendations, reference number [30-32].

Author Contribution Statement

Adam Mohd Khairuddin: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing, Visualization. Siti Armiza Mohd Aris: Writing – review & editing, Validation. Azizul Azizan: Writing – review & editing, Validation. Noor Jannah Zakaria: Writing – review & editing, Validation.