Research Article
AI-Assisted Symptom Classification and Preliminary Disease Prediction System Combining Machine Learning and Large Language Models
@INPROCEEDINGS{10.1007/978-3-032-43712-9_8, author={Chuan-Hung Wu and Nai-Wei Su and Xin-Yu Huang and Lun-Ping Hung}, title={AI-Assisted Symptom Classification and Preliminary Disease Prediction System Combining Machine Learning and Large Language Models}, proceedings={Smart Grid and Internet of Things. 10th EAI International Conference, SGIoT 2026, Taichung, Taiwan, November 21--22, 2026, Proceedings}, proceedings_a={SGIOT}, year={2026}, month={11}, keywords={Large Language Models Machine Learning Disease Prediction Prompt Engineering AI Agent Health Education}, doi={10.1007/978-3-032-43712-9_8} }- Chuan-Hung Wu
Nai-Wei Su
Xin-Yu Huang
Lun-Ping Hung
Year: 2026
AI-Assisted Symptom Classification and Preliminary Disease Prediction System Combining Machine Learning and Large Language Models
SGIOT
Springer
DOI: 10.1007/978-3-032-43712-9_8
Abstract
With the rapid advancement of Large Language Models (LLMs) and generative artificial intelligence, the application of AI in healthcare has attracted growing attention. This study proposes an AI-assisted symptom classification and preliminary disease prediction system integrating machine learning with large language models. The system utilizes a public Disease Symptom Dataset comprising 4,920 records across 41 disease categories to train a Random Forest classification model, and offers a dual-mode input design—a category-organized, tap-to-select symptom interface for the common case, complemented by Ollama with Llama 3.1 as a local LLM that enables natural language symptom interaction when predefined categories do not capture a patient’s description. The prompt design incorporates Constrained Prompting and multi-turn consultation concepts to progressively collect symptom information while reducing hallucination. Experimental results demonstrate that the system achieves 100% classification accuracy on clean inputs and maintains a Top-3 accuracy of 92.78% even when 70% of symptom information is missing, confirming strong robustness under incomplete input conditions within a controlled synthetic dataset. The system further generates plain-language health education recommendations and high-risk alerts through an AI Agent mechanism, establishing a preliminary intelligent healthcare assistance system that more closely reflects actual clinical consultation flows.