A Comparative Study of Symptom-Based Disease Classification Using BERT, SVM and KNN

This paper discusses how Bidirectional Encoder Representations from Transformers (BERT) can be used in developing automated symptom analysis toward enhanced disease diagnosis. BERT's deep contextual understanding is very good at interpreting the complex and very diverse descriptions of symptoms, which made it pick up patterns that are otherwise missed by traditional diagnostic methods. To achieve this, the model was trained on vast medical datasets so that, based on symptom data, it learned to make more precise predictions about the illness. The experiment compares the performance of a set of models: a fine-tuned BERT model, TF -IDF vectorization with KNN classification, and a Support Vector Machine approach. These frameworks are evaluated based on their precision, recall, and Fl-score to identify the most effective solution for disease prediction. The Fine-Tuned BERT -based approach is better than other approaches, both for the improvement of diagnostic accuracy and reducing misdiagnosis, and for providing scalable solutions in real-time clinical decision support.