A Machine Learning Framework for Infectious Disease Prediction via Medical Chatbot Interaction

The integration of medical chatbots with machine learning models for infectious disease prediction remains underexplored, despite its potential to enhance early diagnosis and patient triage. This study introduces a novel framework combining a BERT-based medical chatbot with a hybrid LSTM-Random Forest model to address critical gaps in real-time symptom analysis, diagnostic accuracy, and ethical AI deployment. Unlike existing chatbots limited to general health queries or mental health support, our system employs transformer-based NLP for context-aware patient interactions and integrates temporal symptom analysis with ensemble learning for disease prediction. Additionally, the framework incorporates demographic reweighting to mitigate algorithmic bias and federated learning to preserve patient privacy. Evaluated on the 2023 Global Infectious Disease Dataset (12,354 records), the hybrid model achieved 91.2% accuracy and a 90.1% F1-score, outperforming standalone LSTM (85.6%) and Random Forest (82.3%) models. Bias mitigation reduced accuracy disparities between demographic groups by $4-7 \%$, while federated learning ensured secure data processing. The chatbot demonstrated 82.7% symptom extraction accuracy, surpassing rule-based systems by 12%. These results highlight the framework's potential to advance AI-driven diagnostics by unifying interactive patient engagement, high-precision prediction, and ethical safeguards. The study bridges a critical gap between standalone AI models and patient-facing tools, offering a scalable solution for infectious disease management in diverse healthcare settings.