AI Disease Diagnosis: A Medical Expert System Integrating Traditional Search Algorithms and Machine Learning

AI Disease Diagnosis is an AI-driven medical expert system built with Python and Streamlit, innovatively integrating traditional search algorithms like BFS, A*, and Minimax with machine learning models such as KNN, SVM, and MLP to achieve symptom-driven intelligent disease diagnosis. The project is maintained by simratnvr and published on GitHub (link: https://github.com/simratnvr/ai-disease-diagnosis) on May 26, 2026. Its core innovation lies in combining the symbolic (search algorithms) and connectionist (machine learning) approaches to provide multi-angle analysis for diagnosis.

Medical diagnosis is a complex decision-making process that requires integrating information such as symptoms and medical history. Against the backdrop of uneven distribution of medical resources and limited primary care capabilities, AI-assisted diagnosis has become an important direction. Expert systems were early applications (encoding medical knowledge for reasoning), while machine learning excels at data-driven pattern recognition. The uniqueness of this project lies in integrating these two technical approaches to compensate for the limitations of a single method.

The traditional search algorithm layer includes three classic algorithms:

1. BFS: Uninformed search that traverses all diagnostic paths. It has strong completeness but computational overhead increases with the expansion of the knowledge base, suitable for scenarios with clear associations and limited search space;
2. A Algorithm*: Heuristic search that uses an evaluation function (e.g., symptom-disease association strength, incidence rate) to guide the path, balancing optimal solutions and efficiency;
3. Minimax Algorithm: A game theory approach that treats diagnosis as a game against diseases, providing a formal framework for handling uncertainty.

The machine learning model layer integrates three classification algorithms:

1. KNN: Instance-based learning that makes decisions via voting from similar historical cases. It is intuitive and easy to understand, requires no training, and adapts to multi-classification;
2. SVM: Finds the optimal decision boundary, excels at high-dimensional sparse data (e.g., symptom combinations), and may use multi-class variants to handle disease differentiation;
3. MLP: A basic form of deep learning that captures complex non-linear mappings between symptoms and enables high-order interaction effects.

The project chooses Python as the development language, leveraging its data science and machine learning ecosystem; Streamlit as the web framework to quickly build an interactive demonstration system. This lightweight technology stack is suitable for academic research and prototype verification, facilitating rapid iteration and multi-party validation.

Application Value:

• Educational research: A complete end-to-end example to help learners understand the application of AI technology in healthcare and the construction of Streamlit interactions;
• Technical exploration: The hybrid architecture inspires the complementarity of different AI paradigms, leveraging their strengths (interpretability of symbolic reasoning and pattern recognition of data-driven approaches). Limitations:
• Only a technical demonstration/educational example, lacking clinical validation and regulatory approval, and cannot replace professional doctors;
• The scale and quality of the disease knowledge base and training data are not detailed, affecting accuracy assessment.

Future Directions:

• Technical improvements: Introduce Transformers and graph neural networks to capture complex relationships; integrate medical knowledge graphs to enhance interpretability; support natural language symptom input;
• Application evolution: Build specialized systems for specific disease areas, collaborate with medical institutions to obtain real data for verification and optimization. Summary: The project demonstrates the potential of AI in medical diagnosis. Although it is still far from clinical application, it provides a reference for research and learning. In the future, with data accumulation, algorithm progress, and regulatory improvement, AI-assisted diagnosis will enhance healthcare accessibility and quality.